File size: 72,156 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
{
    "paper_id": "W09-0203",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T06:39:57.240333Z"
    },
    "title": "Unsupervised Classification with Dependency Based Word Spaces",
    "authors": [
        {
            "first": "Klaus",
            "middle": [],
            "last": "Rothenh\u00e4usler",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "University of Stuttgart",
                "location": {
                    "settlement": "Stuttgart",
                    "country": "Germany"
                }
            },
            "email": "[email protected]"
        },
        {
            "first": "Hinrich",
            "middle": [],
            "last": "Sch\u00fctze",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "University of Stuttgart",
                "location": {
                    "settlement": "Stuttgart",
                    "country": "Germany"
                }
            },
            "email": "[email protected]"
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "We present the results of clustering experiments with a number of different evaluation sets using dependency based word spaces. Contrary to previous results we found a clear advantage using a parsed corpus over word spaces constructed with the help of simple patterns. We achieve considerable gains in performance over these spaces ranging between 9 and 13% in absolute terms of cluster purity.",
    "pdf_parse": {
        "paper_id": "W09-0203",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "We present the results of clustering experiments with a number of different evaluation sets using dependency based word spaces. Contrary to previous results we found a clear advantage using a parsed corpus over word spaces constructed with the help of simple patterns. We achieve considerable gains in performance over these spaces ranging between 9 and 13% in absolute terms of cluster purity.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "Word space models have become a mainstay in the automatic acquisition of lexical semantic knowledge. The computation of semantic relatedness of two words in such models is based on their distributional similarity. The most crucial way in which such models differ is the definition of distributional similarity: In a regular word space model the observed distribution concerns the immediate neighbours of a word within a predefined window to the left and right (Sch\u00fctze, 1992; Sahlgren, 2006) . Early on in the development as an alternative models were proposed that relied on the similarity of the distribution of syntactic relations (Hindle, 1990; Pad\u00f3 and Lapata, 2007) . More recently the distribution of the occurrence within simple patterns defined in the form of regular expressions that are supposed to capture explicit semantic relations was explored as the basis of distributional similarity (Almuhareb and Poesio, 2004) .",
                "cite_spans": [
                    {
                        "start": 460,
                        "end": 475,
                        "text": "(Sch\u00fctze, 1992;",
                        "ref_id": "BIBREF15"
                    },
                    {
                        "start": 476,
                        "end": 491,
                        "text": "Sahlgren, 2006)",
                        "ref_id": "BIBREF14"
                    },
                    {
                        "start": 634,
                        "end": 648,
                        "text": "(Hindle, 1990;",
                        "ref_id": "BIBREF9"
                    },
                    {
                        "start": 649,
                        "end": 671,
                        "text": "Pad\u00f3 and Lapata, 2007)",
                        "ref_id": "BIBREF12"
                    },
                    {
                        "start": 901,
                        "end": 929,
                        "text": "(Almuhareb and Poesio, 2004)",
                        "ref_id": "BIBREF0"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Whereas dependency based semantic spaces have been shown to surpass other word space models for a number of problems (Pad\u00f3 and Lapata, 2007; Lin, 1998) , for the task of categorisation simple pattern based spaces have been shown to perform equally good if not better . We want to show that dependency based spaces also fare better in these tasks if the dependency relations used are selected reasonably. At the same time we want to show that such a system can be built with freely available components and without the need to rely on the index of a proprietary search engine vendor.",
                "cite_spans": [
                    {
                        "start": 117,
                        "end": 140,
                        "text": "(Pad\u00f3 and Lapata, 2007;",
                        "ref_id": "BIBREF12"
                    },
                    {
                        "start": 141,
                        "end": 151,
                        "text": "Lin, 1998)",
                        "ref_id": "BIBREF11"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "We propose to use the web acquired data of the ukWaC (Ferraresi et al., 2008) , which is huge but still manageable and comes in a pre-cleaned version with HTML markup removed. It can easily be fed into a parser like MiniPar which allows for the subsequent extraction of dependency relations of different types and complexity. In particular we work with dependency paths that can reach beyond direct dependencies as opposed to Lin (1998) but in the line of Pado and Lapata (2007) . In contrast to the latter, however, different paths that end in the same word are not generally mapped to the same dimension in our model. A path in a dependency graph can pass through several nodes and encompass different relations.",
                "cite_spans": [
                    {
                        "start": 53,
                        "end": 77,
                        "text": "(Ferraresi et al., 2008)",
                        "ref_id": "BIBREF7"
                    },
                    {
                        "start": 426,
                        "end": 436,
                        "text": "Lin (1998)",
                        "ref_id": "BIBREF11"
                    },
                    {
                        "start": 456,
                        "end": 478,
                        "text": "Pado and Lapata (2007)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "We experimented with two sets of nouns previously used in the literature for word clustering. The nouns in both sets are taken from a number of different WordNet categories. Hence, the task consists in clustering together the words from the same category. By keeping the clustering algorithm constant, differences in performance can be attributed to the differences of the word representations.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "The next section provides a formal description of our word space model. Section 3 reports on our clustering experiments with two sets of concepts used previously to evaluate the categorisation abilities of word spaces. Section 4 discusses these re-sults and draws some conclusions.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "We follow the formalisation and terminology developed in Pado and Lapata (2007) according to which a dependency based space is determined by the sets of its basis elements B and targets T that form a matrix M = B \u00d7 T , a similarity function S that assigns a real-valued similarity measure to pairs of elements from T , the association measure A that captures the strength of the relation between a target and a basis element, the context selection function cont, the basis mapping function \u00b5 and the path value function v. Our set of targets is always a subset of the lemmas output by MiniPar. The remaining elements are defined in this section. We use \u03c0 to denote a path in a dependency graph which is conceived of as an undirected graph for this purpose. So, in general a dependency path has an upward and downward part where one can have length zero. All the paths used to define the contexts for target words are anchored there, i.e. they start from the target.",
                "cite_spans": [
                    {
                        "start": 44,
                        "end": 79,
                        "text": "developed in Pado and Lapata (2007)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Word Space Construction",
                "sec_num": "2"
            },
            {
                "text": "In choosing the context definitions that determine what dependency paths are used in the construction of the word vectors, we oriented ourselves at the sets proposed in Pado and Lapata (2007) . As Pado and Lapata (2007) achieved their best results with it we started from their medium sized set of context definitions, from which we extracted the appropriate ones for our experiments and added some that seemed to make sense for our purposes: As our evaluation sets consist entirely of nouns, we used only context definitions that start at a noun. Thereby we can ensure that only nominal uses are recorded in a word vector if a target word can have different parts of speech. The complete set of dependency relations our context selection function cont comprises is given in Figure 1 along with an example for each.",
                "cite_spans": [
                    {
                        "start": 178,
                        "end": 191,
                        "text": "Lapata (2007)",
                        "ref_id": "BIBREF12"
                    },
                    {
                        "start": 197,
                        "end": 219,
                        "text": "Pado and Lapata (2007)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [
                    {
                        "start": 775,
                        "end": 783,
                        "text": "Figure 1",
                        "ref_id": "FIGREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Word Space Construction",
                "sec_num": "2"
            },
            {
                "text": "We only chose paths that end in an open word class assuming that they are more informative about the meaning of a target word. Paths ending in a preposition for instance, as used by Pado and Lapata (2007) , were not considered. For the same reason we implemented a simple stop word filter that discards paths ending in a pronoun, which are assigned the tag N by MiniPar just like any other noun.",
                "cite_spans": [
                    {
                        "start": 182,
                        "end": 204,
                        "text": "Pado and Lapata (2007)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Word Space Construction",
                "sec_num": "2"
            },
            {
                "text": "On the other hand we added the relation between a prepositional complement and the noun it modifies (appearing as relation IX in Figure 1 ) as a close approximation of the pattern used by (Almuhareb and Poesio, 2004) to identify attributes of a concept as detailed in the next section. Path specifications X and XI are also additions we made that are thought to gather additional attribute values to the ones already covered by III.",
                "cite_spans": [
                    {
                        "start": 203,
                        "end": 216,
                        "text": "Poesio, 2004)",
                        "ref_id": "BIBREF0"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 129,
                        "end": 137,
                        "text": "Figure 1",
                        "ref_id": "FIGREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Word Space Construction",
                "sec_num": "2"
            },
            {
                "text": "As a basis mapping function \u00b5 we used a generalisation of the one used by Grefenstette (1994) and Lin (1998) . They map a dependency between two words to a pair consisting of the relation label l and the end word of the dependency end(\u03c0). As we use paths that span more than a single relation, this approach is not directly applicable to our setup. Instead we use a mapping function that maps a path to the sequence of edge labels through which it passes combined with the end word:",
                "cite_spans": [
                    {
                        "start": 74,
                        "end": 93,
                        "text": "Grefenstette (1994)",
                        "ref_id": "BIBREF8"
                    },
                    {
                        "start": 98,
                        "end": 108,
                        "text": "Lin (1998)",
                        "ref_id": "BIBREF11"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Word Space Construction",
                "sec_num": "2"
            },
            {
                "text": "\u00b5(\u03c0) = (l(\u03c0), end(\u03c0))",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Word Space Construction",
                "sec_num": "2"
            },
            {
                "text": "where l(\u2022) is a labelling function that returns the sequence of edge labels for a given path. With this basis mapping function the nodes or words respectively through which a path passes are all neglected except for the node where the path ends. So, for the noun human the sequence human and mouse genome as well as the sequence human and chimpanzee genome increase the count for the same basis element :N:conj:N: * :N:nn:N:genome. Here we use a path notation of the general form:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Word Space Construction",
                "sec_num": "2"
            },
            {
                "text": "(: POS : rel : POS : {word, * }) n",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Word Space Construction",
                "sec_num": "2"
            },
            {
                "text": "where POS is a part of speech, rel a relation and word a node label, i.e. a lemma, all as produced by MiniPar. The length of a path is determined by n and the asterisk ( * ) indicates that a node label is ignored by the basis mapping function.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Word Space Construction",
                "sec_num": "2"
            },
            {
                "text": "As an alternative we experimented with a lexical basis mapping function that maps a path to its end word:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Word Space Construction",
                "sec_num": "2"
            },
            {
                "text": "\u00b5(\u03c0) = end(\u03c0)",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Word Space Construction",
                "sec_num": "2"
            },
            {
                "text": "This reduces the number of dimensions considerably and yields semantic spaces that are similar to window based word spaces. As this mapping function consistently delivered worse results, we dropped it from our evaluation.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Word Space Construction",
                "sec_num": "2"
            },
            {
                "text": "Considering that (Pad\u00f3 and Lapata, 2007) only reported very small differences for different path valuation functions, we only used a constant valuation of paths: Thus, an occurrence of any path, irrespective of length or grammatical relations that are involved, increases the count of the respective basis element by one. We implemented three different association functions, A, to transform the raw frequency counts and weight the influence of the different cooccurrences. We worked with an implementation of the log likelihood ratio (g-Score) as proposed by Dunning (1993) and two variants of the t-score, one considering all values (t-score) and one where only positive values (t-score + ) are kept following the results of Curran and Moens (2002) . We also experimented with different frequency cutoffs removing dimensions that occur very frequently or very rarely.",
                "cite_spans": [
                    {
                        "start": 17,
                        "end": 40,
                        "text": "(Pad\u00f3 and Lapata, 2007)",
                        "ref_id": "BIBREF12"
                    },
                    {
                        "start": 560,
                        "end": 574,
                        "text": "Dunning (1993)",
                        "ref_id": "BIBREF6"
                    },
                    {
                        "start": 727,
                        "end": 750,
                        "text": "Curran and Moens (2002)",
                        "ref_id": "BIBREF5"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Word Space Construction",
                "sec_num": "2"
            },
            {
                "text": "v const (\u03c0) = 1 (I)",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Word Space Construction",
                "sec_num": "2"
            },
            {
                "text": "For all our experiments we used the ukWaC corpus 1 to construct the word spaces, which was parsed using MiniPar. The latter provides lemma information, which we used as possible target and context words. The word vectors we built from this data were represented as pseudo documents in an inverted index. To our knowledge the experiments described in this paper are the first to work with a completely parsed version of the ukWaC.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Evaluation",
                "sec_num": "3"
            },
            {
                "text": "For the evaluation the word vectors for the test sets were clustered into a predefined number of clusters corresponding to the number of concept classes from which the words were drawn. All experiments were conducted with the CLUTO toolkit (Karypis, 2003) using the repeated bisections clustering algorithm with global optimisation and the cosine as a distance measure to maintain comparability with related work, e.g. .",
                "cite_spans": [
                    {
                        "start": 240,
                        "end": 255,
                        "text": "(Karypis, 2003)",
                        "ref_id": "BIBREF10"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Evaluation",
                "sec_num": "3"
            },
            {
                "text": "As the main evaluation measure we used purity for the whole set as supplied by CLUTO. For a clustering solution \u2126 of n clusters and a set of classes C, purity can be defined as:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Evaluation",
                "sec_num": "3"
            },
            {
                "text": "purity(\u2126,C) = 1 n \u2211 k max j |\u03c9 k \u2229 c j |",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Evaluation",
                "sec_num": "3"
            },
            {
                "text": "where \u03c9 k denotes the set of terms in a cluster and c j the set of terms in a class. This aggregate measure of purity corresponds to the weighted sum of purities for the individual clusters, which is defined as the ratio of items in a cluster that belong to the majority class. The results for the two test 1 http://wacky.sslmit.unibo.it sets we used are described in the following two subsections.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Evaluation",
                "sec_num": "3"
            },
            {
                "text": "The first set we worked with was introduced by Almuhareb and Poesio (2004) A match for the concept [HUMAN] would be the dignity of the human is, which yields dignity as an attribute. These patterns were translated into queries and submitted to the Google 2 search engine. We compare our dependency based spaces with the results achieved with the pattern based approach in Table 1 Table 1 : Categorisation results for the 214 concepts and 13 classes proposed in Almuhareb and Poesio (2004) , which is also the source of the result for the pattern based space. They only used t-score + . The numbers given are the best accuracies achieved under the different settings.",
                "cite_spans": [
                    {
                        "start": 47,
                        "end": 74,
                        "text": "Almuhareb and Poesio (2004)",
                        "ref_id": "BIBREF0"
                    },
                    {
                        "start": 475,
                        "end": 488,
                        "text": "Poesio (2004)",
                        "ref_id": "BIBREF0"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 372,
                        "end": 379,
                        "text": "Table 1",
                        "ref_id": null
                    },
                    {
                        "start": 380,
                        "end": 387,
                        "text": "Table 1",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Results for 214 nouns from Almuhareb and Poesio (2004)",
                "sec_num": "3.1"
            },
            {
                "text": "For the window based space we used the best performing in a free association task with a window size of six words to each side and all the words that appeared at least two times as dimensions ignoring stop words. The effective dimensionality of the so built word vectors is 417 837.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Results for 214 nouns from Almuhareb and Poesio (2004)",
                "sec_num": "3.1"
            },
            {
                "text": "The results for the dependency based spaces were built by selecting all paths without any frequency thresholds which resulted in a set of 767 119 dimensions.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Results for 214 nouns from Almuhareb and Poesio (2004)",
                "sec_num": "3.1"
            },
            {
                "text": "As can be seen, both window and dependency based spaces exceed the pattern based space for certain association measures. But the dependency space also has a clear advantage over the window based space. In particular the t-score + measure yields very good results. In contrast the g-score offers the worst results with the t-score retaining negative values somewhere in between. For our further experiments we hence used the t-score + association measure.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Results for 214 nouns from Almuhareb and Poesio (2004)",
                "sec_num": "3.1"
            },
            {
                "text": "We ran a number of experiments to quantify the impact the different kinds of paths have on the clustering result. We first built spaces using only a single kind of path to find out how good each performs on its own. The result can be found in Table 2 . For some of the words in the evaluation set no contexts could be found when only one of the two most complex context specifications (X), (XI) was used or when the context was reduced to the genitive relation (IV). Apart from that the results suggest that even a single type of relation on its own can prove highly effective. Especially the conjunctive relation (VI) performs very well with a purity value of 95.3%. Table 3 : Clustering results for spaces with one context specification removed.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 243,
                        "end": 250,
                        "text": "Table 2",
                        "ref_id": "TABREF4"
                    },
                    {
                        "start": 668,
                        "end": 675,
                        "text": "Table 3",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Further Analysis",
                "sec_num": "3.1.1"
            },
            {
                "text": "To further clarify the role of the different kinds of contexts, we ran the experiment with word spaces where we removed each one of the twelve context specifications in turn. The results as given in Table 3 are a bit astonishing at first sight: Only the removal of the conjunctive relation actually leads to a decrease in performance. All the other contexts seem to be either redundant -with performance staying the same when they are removed -or even harmful -with performance increasing once they are removed. Having observed this, we tried to remove further context specifications and surprisingly found that the best performance of 98.1% can be reached by only including the conjunction (VI) and the object (II) relations. The dimensionality of these vectors is only a fraction of the original ones with 100 597.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 199,
                        "end": 206,
                        "text": "Table 3",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Further Analysis",
                "sec_num": "3.1.1"
            },
            {
                "text": "The result for the best performing dependency based space listed in the table is almost perfect. Having a closer look at the results reveals that in fact only four words are put into a wrong cluster. These words are: lounge, pain, mouse, oyster.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Further Analysis",
                "sec_num": "3.1.1"
            },
            {
                "text": "The first is classified as [BUILDING] instead of [FURNITURE] . In the case of lounge the misclassification seems to be attributable to the ambiguity of the word which can either denote a piece of furniture or a waiting room. The latter is apparently the more prominent sense in the data. In this usage the word often appears in conjunctions with room or hotel just like restaurant, inn or clubhouse.",
                "cite_spans": [
                    {
                        "start": 49,
                        "end": 60,
                        "text": "[FURNITURE]",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Further Analysis",
                "sec_num": "3.1.1"
            },
            {
                "text": "Pain is misclassified as an [ILLNESS] instead of a [FEELING] which is at least a close miss. The misclassification of mouse as a [BODY PART] seems rather odd on the other hand. The reason for it becomes apparent when looking at the most descriptive and discriminating features of the [BODY PART] cluster: In both lists the highest in the ranking is the dimension :N:mod:A:left, i.e. left as an adjectival modifier of the word in question. The prominence of this particular modification is of course due to the fact that a lot of body parts come in pairs and that the members of these pairs are commonly identified by assigning them to the left or right half of the body. Certainly, the word mouse enters this cluster not through its sense of mouse 1 as an animal but rather through its sense of mouse 2 as a piece of computer equipment that has two buttons, which are also referred to as the left and right one. Unfortunately, MiniPar frequently resolves left in a wrong way as a modifier of mouse instead of button.",
                "cite_spans": [
                    {
                        "start": 51,
                        "end": 60,
                        "text": "[FEELING]",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Further Analysis",
                "sec_num": "3.1.1"
            },
            {
                "text": "Finally for oyster which is put into the [EDIBLE FRUIT] instead of the [ANIMAL] cluster it is conspicuous that oyster is the only sea animal in the evaluation set and consequently it rarely occurs in conjunctions with the other animals. Conjunctions, however, seem to be the most important features for defining all the clusters. Additionally oyster scores low on a lot of dimensions that are typical for a big number of the members of the animal cluster, e.g. :N:obj:V:kill.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Further Analysis",
                "sec_num": "3.1.1"
            },
            {
                "text": "In Poesio and Almuhareb (2005a) a larger evaluation set is introduced that comprises 402 nouns sampled from the hierarchies under the 21 unique beginners in WordNet. The words were also chosen so that candidates from different frequency bands and different levels of ambiguity were represented. Further results using this set are reported in . The best result was obtained with the attribute pattern alone and filtering to include only nouns. We tried to assemble word vectors with the same patterns based on the ukWaC corpus. But even if we included both patterns, we were only able to construct vectors for 363 of the 402 words. For 118 of them the number of occurrences, on which they were based, was less than ten. This gives an impression of the size of the index that is necessary for such an approach. To date such an immense amount of data is only available through proprietary search engine providers. This makes a system dependant upon the availability of an API of such a vendor. In fact the version of the Google API on which the original experiments relied has since been axed. Our approach circumvents such problems. We ran analogous experiments to the ones described in the previous section on this evaluation set, now producing 21 clusters. The results given in Table 4 are for a dependency space without any frequency thresholds and the complete set of context specifications as defined above. The settings for the window based space were also the same (6 words to each side). Again the results achieved with the t-score + association were clearly superior to the others and were used in all the following experiments. Unsurprisingly, for this more difficult task the performance is not as good as for the smaller set but nevertheless the superiority of the dependency based space is clearly visible with an absolute increase in cluster purity of 8.2% compared with the pattern based space. Table 4 : Categorisation results for the 402 concepts and 21 classes proposed in Almuhareb and Poesio (2005a) which is also the source of the result for the pattern based space. The numbers given are the best accuracies achieved under the different settings.",
                "cite_spans": [
                    {
                        "start": 3,
                        "end": 31,
                        "text": "Poesio and Almuhareb (2005a)",
                        "ref_id": "BIBREF2"
                    },
                    {
                        "start": 2003,
                        "end": 2017,
                        "text": "Poesio (2005a)",
                        "ref_id": "BIBREF2"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 1278,
                        "end": 1285,
                        "text": "Table 4",
                        "ref_id": null
                    },
                    {
                        "start": 1908,
                        "end": 1915,
                        "text": "Table 4",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Results for 402 words from Almuhareb and Poesio (2005a)",
                "sec_num": "3.2"
            },
            {
                "text": "Again we ran further experiments to determine the impact of the different kinds of relations. The removal of any single context specification leads to a performance drop with this evaluation set. The smallest decrease is observed when removing context specification XII. However, as we had seen in the previous experiment with the smaller set that only two context specifications suffice to reach peak performance, we conducted another experiment where we started from the best performing space constructed from a single context specification (the conjunction relation, VI) and successively added the specification that led to the biggest performance gain. The crucial results are majority class concepts solid tetrahedron, salient, ring, ovoid, octahedron, knob, icosahedron, fluting, dome, dodecahedron, cylinder, cuboid, cube, crinkle, concavity, samba, coco, nonce, divan, ball, stitch, floater, trove, hoard, mouse time yesteryear, yesterday, tonight, tomorrow, today, quaternary, period, moment, hereafter, gestation, future, epoch, day, date, aeon, stretch, snap, throb, straddle, nap motivation wanderlust, urge, superego, obsession, morality, mania, life, impulse, ethics, dynamic, conscience, compulsion, plasticity, opinion, acceptance, sensitivity, desire, interest assets wager, taxation, quota, profit, payoff, mortgage, investment, income, gain, fund, credit, capital, allotment, allocation, possession, inducement, incentive, disincentive, deterrence, share, sequestrian, cheque, check, bond, tailor district village, town, sultanate, suburb, state, shire, seafront, riverside, prefecture, parish, metropolis, land, kingdom, county, country, city, canton, borough, borderland, anchorage, tribe, nation, house, fen, cordoba, faro legal document treaty, statute, rescript, obligation, licence, law, draft, decree, convention, constitution, bill, assignment, commencement, extension, incitement, caliphate, clemency, venture, dispensation physical property weight, visibility, temperature, radius, poundage, momentum, mass, length, diameter, deflection, taper, indentation, droop, corner, concavity social unit troop , team, platoon, office, legion, league, household, family, department, confederacy, company, committee, club, bureau, brigade, branch, agency atmospheric phenomenon wind, typhoon, tornado, thunderstorm, snowfall, shower, sandstorm, rainstorm, lightning, hurricane, fog, drizzle, cyclone, crosswind, cloudburst, cloud, blast, aurora, airstream, glow social occasion wedding, rededication, prom, pageantry, inaugural, graduation, funeral, fundraiser, fiesta, fete, feast, enthronement, dance, coronation, commemoration, ceremony, celebration, occasion, raffle, beano monetary unit zloty, yuan, shilling, rupee, rouble, pound, peso, penny, lira, guilder, franc, escudo, drachma, dollar, dirham, dinar, cent tree sycamore, sapling, rowan, pine, palm, oak, mangrove, jacaranda, hornbeam, conifer, cinchona, casuarina, acacia, riel chemical element zinc, titanium, silver, potassium, platinum, oxygen, nitrogen, neon, magnesium, lithium, iron, hydrogen, helium, germanium, copper, charcoal, carbon, calcium, cadmium, bismuth, aluminium, gold illness smallpox, plague, meningitis, malnutrition, leukemia, hepatitis, glaucoma, flu, eczema, diabetes, cirrhosis, cholera, cancer, asthma, arthritis, anthrax, acne, menopause feeling wonder, shame, sadness, pleasure, passion, love, joy, happiness, fear, anger, heaviness, coolness, torment, tenderness, suffering, stinging vehicle van, truck, ship, rocket, pickup, motorcycle, helicopter, cruiser, car, boat, bicycle, automobile, airplane, aircraft, jag creator producer, photographer, painter, originator, musician, manufacturer, maker, inventor, farmer, developer, designer, craftsman, constructor, builder, artist, architect, motivator pain toothache, soreness, sting, soreness, sciatica, neuralgia, migraine, lumbago, headache, earache, burn, bellyache, backache, ache, rheumatism, pain animal zebra, turtle, tiger, sheep, rat, puppy, monkey, lion, kitten, horse, elephant, dog, deer, cow, cat, camel, bull, bear game whist, volleyball, tennis, softball, soccer, rugby, lotto, keno, handball, golf, football, curling, chess, bowling, basketball, baccarat, twister edible fruit watermelon, strawberry, pineapple, pear, peach, orange, olive, melon, mango, lemon, kiwi, grape, cherry, berry, banana, apple, oyster, walnut, pistachio, mandarin, lime, fig, chestnut Figure 2: Optimal clustering for large evaluation set.",
                "cite_spans": [
                    {
                        "start": 711,
                        "end": 4434,
                        "text": "tetrahedron, salient, ring, ovoid, octahedron, knob, icosahedron, fluting, dome, dodecahedron, cylinder, cuboid, cube, crinkle, concavity, samba, coco, nonce, divan, ball, stitch, floater, trove, hoard, mouse time yesteryear, yesterday, tonight, tomorrow, today, quaternary, period, moment, hereafter, gestation, future, epoch, day, date, aeon, stretch, snap, throb, straddle, nap motivation wanderlust, urge, superego, obsession, morality, mania, life, impulse, ethics, dynamic, conscience, compulsion, plasticity, opinion, acceptance, sensitivity, desire, interest assets wager, taxation, quota, profit, payoff, mortgage, investment, income, gain, fund, credit, capital, allotment, allocation, possession, inducement, incentive, disincentive, deterrence, share, sequestrian, cheque, check, bond, tailor district village, town, sultanate, suburb, state, shire, seafront, riverside, prefecture, parish, metropolis, land, kingdom, county, country, city, canton, borough, borderland, anchorage, tribe, nation, house, fen, cordoba, faro legal document treaty, statute, rescript, obligation, licence, law, draft, decree, convention, constitution, bill, assignment, commencement, extension, incitement, caliphate, clemency, venture, dispensation physical property weight, visibility, temperature, radius, poundage, momentum, mass, length, diameter, deflection, taper, indentation, droop, corner, concavity social unit troop , team, platoon, office, legion, league, household, family, department, confederacy, company, committee, club, bureau, brigade, branch, agency atmospheric phenomenon wind, typhoon, tornado, thunderstorm, snowfall, shower, sandstorm, rainstorm, lightning, hurricane, fog, drizzle, cyclone, crosswind, cloudburst, cloud, blast, aurora, airstream, glow social occasion wedding, rededication, prom, pageantry, inaugural, graduation, funeral, fundraiser, fiesta, fete, feast, enthronement, dance, coronation, commemoration, ceremony, celebration, occasion, raffle, beano monetary unit zloty, yuan, shilling, rupee, rouble, pound, peso, penny, lira, guilder, franc, escudo, drachma, dollar, dirham, dinar, cent tree sycamore, sapling, rowan, pine, palm, oak, mangrove, jacaranda, hornbeam, conifer, cinchona, casuarina, acacia, riel chemical element zinc, titanium, silver, potassium, platinum, oxygen, nitrogen, neon, magnesium, lithium, iron, hydrogen, helium, germanium, copper, charcoal, carbon, calcium, cadmium, bismuth, aluminium, gold illness smallpox, plague, meningitis, malnutrition, leukemia, hepatitis, glaucoma, flu, eczema, diabetes, cirrhosis, cholera, cancer, asthma, arthritis, anthrax, acne, menopause feeling wonder, shame, sadness, pleasure, passion, love, joy, happiness, fear, anger, heaviness, coolness, torment, tenderness, suffering, stinging vehicle van, truck, ship, rocket, pickup, motorcycle, helicopter, cruiser, car, boat, bicycle, automobile, airplane, aircraft, jag creator producer, photographer, painter, originator, musician, manufacturer, maker, inventor, farmer, developer, designer, craftsman, constructor, builder, artist, architect, motivator pain toothache, soreness, sting, soreness, sciatica, neuralgia, migraine, lumbago, headache, earache, burn, bellyache, backache, ache, rheumatism, pain animal zebra, turtle, tiger, sheep, rat, puppy, monkey, lion, kitten, horse, elephant, dog, deer, cow, cat, camel, bull, bear game whist, volleyball, tennis, softball, soccer, rugby, lotto, keno, handball, golf, football, curling, chess, bowling, basketball, baccarat, twister edible fruit watermelon, strawberry, pineapple, pear, peach, orange, olive, melon, mango, lemon, kiwi, grape, cherry, berry, banana, apple, oyster, walnut, pistachio, mandarin, lime, fig, chestnut",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Further Analysis",
                "sec_num": "3.2.1"
            },
            {
                "text": "contexts used purity (VI) 73.4% (VI), (II) 76.6% (VI), (II), (III) 80.1% Table 5 : Clustering the larger evaluation set with an increasing number of context specifications.",
                "cite_spans": [
                    {
                        "start": 38,
                        "end": 42,
                        "text": "(II)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [
                    {
                        "start": 73,
                        "end": 80,
                        "text": "Table 5",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Further Analysis",
                "sec_num": "3.2.1"
            },
            {
                "text": "given in Table 5 . As can be seen the object relation is added first again. This time though the inclusion of adjectival modification brings another performance increase which is even one per cent above the result for the space built from all possible relations. The addition of any further contexts consistently degrades performance. The clustering solution thus produced is given in Figure 2 . From the 1 872 698 dimension used in the original space only 341 214 are retained.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 9,
                        "end": 16,
                        "text": "Table 5",
                        "ref_id": null
                    },
                    {
                        "start": 385,
                        "end": 393,
                        "text": "Figure 2",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Further Analysis",
                "sec_num": "3.2.1"
            },
            {
                "text": "Our results are counterintuitive at first sight as it could be expected that a larger number of different contexts would increase performance. Instead we see the best performance with only a very lim-ited set of possible contexts. We suspect that this behaviour is due to a large amount of correlation between the different kinds of contexts. The addition of further contexts beyond a certain point therefore has no positive effect. As an indication for this it might be noticed that the three context specifications that yield the best result for the 402 word set comprise relations with the three main open word classes. It is to be expected that they contribute orthogonal information that covers central dimensions of meaning. The slight decrease in performance that can be observed when further contexts are added is probably due to chance fluctuations and almost certainly not significant; with significance being hard to determine for any of the results.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Discussion and Conclusion",
                "sec_num": "4"
            },
            {
                "text": "However, it is obviously necessary to cover a basic variety of features. Patterns which are used to explicitly track semantic relations on the textual surface seem to be too restrictive. Information accessible from co-occurring verbs for example is completely lost. In a regular window based word space such information is retained and its performance is competitive with a pattern based approach. This method is obviously too liberal, though, if compared to the dependency spaces.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Discussion and Conclusion",
                "sec_num": "4"
            },
            {
                "text": "In general we were able to show that semantic spaces are obviously able to capture categorical knowledge about concepts best when they are built from a syntactically annotated source. This is true even if the context specification used is not the most parsimonious. The problem of determining the right set of contexts is therefore rather an optimisation issue than a question of using dependency based spaces or not. It is a considerable one, though, as computations are much cheaper with vectors of reduced dimensionality, of course.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Discussion and Conclusion",
                "sec_num": "4"
            },
            {
                "text": "For the categorisation task the inclusion of more complex relations reaching over several dependencies does not seem to be helpful considering they can all be dropped without a decrease in performance. As Pado and Lapata (2007) reached better results in their experiments with a broader set of context specifications we conclude that the selection of the kinds of context to include when constructing a word space depends largely on the task at hand.",
                "cite_spans": [
                    {
                        "start": 205,
                        "end": 227,
                        "text": "Pado and Lapata (2007)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Discussion and Conclusion",
                "sec_num": "4"
            },
            {
                "text": "http://www.google.com",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            }
        ],
        "back_matter": [],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "Attributebased and value-based clustering: An evaluation",
                "authors": [
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Almuhareb",
                        "suffix": ""
                    },
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Poesio",
                        "suffix": ""
                    }
                ],
                "year": 2004,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "A. Almuhareb and M. Poesio. 2004. Attribute- based and value-based clustering: An evaluation.",
                "links": null
            },
            "BIBREF1": {
                "ref_id": "b1",
                "title": "Proceedings of EMNLP 2004",
                "authors": [],
                "year": null,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "158--165",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "In Dekang Lin and Dekai Wu, editors, Proceedings of EMNLP 2004, pages 158-165, Barcelona, Spain, July. Association for Computational Linguistics.",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "Concept learning and categorization from the web",
                "authors": [
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Poesio",
                        "suffix": ""
                    },
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Almuhareb",
                        "suffix": ""
                    }
                ],
                "year": 2005,
                "venue": "Proceedings of CogSci2005 -XXVII Annual Conference of the Cognitive Science Society",
                "volume": "",
                "issue": "",
                "pages": "103--108",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "M. Poesio and A. Almuhareb. 2005a. Concept learn- ing and categorization from the web. In Proceedings of CogSci2005 -XXVII Annual Conference of the Cognitive Science Society, pages 103-108, Stresa, Italy.",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "Finding attributes in the web using a parser",
                "authors": [
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Almuhareb",
                        "suffix": ""
                    },
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Poesio",
                        "suffix": ""
                    }
                ],
                "year": 2005,
                "venue": "Proceedings of Corpus Linguistics",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "A. Almuhareb and M. Poesio. 2005b. Finding at- tributes in the web using a parser. In Proceedings of Corpus Linguistics, Birmingham.",
                "links": null
            },
            "BIBREF5": {
                "ref_id": "b5",
                "title": "Improvements in automatic thesaurus extraction",
                "authors": [
                    {
                        "first": "J",
                        "middle": [
                            "R"
                        ],
                        "last": "Curran",
                        "suffix": ""
                    },
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Moens",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "Proceedings of the ACL-02 workshop on Unsupervised lexical acquisition",
                "volume": "",
                "issue": "",
                "pages": "59--66",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "J. R. Curran and M. Moens. 2002. Improvements in automatic thesaurus extraction. In Proceedings of the ACL-02 workshop on Unsupervised lexical ac- quisition, pages 59-66, Morristown, NJ, USA. As- sociation for Computational Linguistics.",
                "links": null
            },
            "BIBREF6": {
                "ref_id": "b6",
                "title": "Accurate methods for the statistics of surprise and coincidence",
                "authors": [
                    {
                        "first": "T",
                        "middle": [],
                        "last": "Dunning",
                        "suffix": ""
                    }
                ],
                "year": 1993,
                "venue": "Computational Linguistics",
                "volume": "19",
                "issue": "1",
                "pages": "61--74",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "T. Dunning. 1993. Accurate methods for the statistics of surprise and coincidence. Computational Lin- guistics, 19(1):61-74.",
                "links": null
            },
            "BIBREF7": {
                "ref_id": "b7",
                "title": "Introducing and evaluating ukwac, a very large web-derived corpus of english",
                "authors": [
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Ferraresi",
                        "suffix": ""
                    },
                    {
                        "first": "E",
                        "middle": [],
                        "last": "Zanchetta",
                        "suffix": ""
                    },
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Baroni",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Bernardini",
                        "suffix": ""
                    }
                ],
                "year": 2008,
                "venue": "Proceedings of the WAC4 Workshop at LREC",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "A. Ferraresi, E. Zanchetta, M. Baroni, and S. Bernar- dini. 2008. Introducing and evaluating ukwac, a very large web-derived corpus of english. In Pro- ceedings of the WAC4 Workshop at LREC 2008.",
                "links": null
            },
            "BIBREF8": {
                "ref_id": "b8",
                "title": "Explorations in Automatic Thesaurus Discovery",
                "authors": [
                    {
                        "first": "G",
                        "middle": [],
                        "last": "Grefenstette",
                        "suffix": ""
                    }
                ],
                "year": 1994,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "G. Grefenstette. 1994. Explorations in Automatic The- saurus Discovery. Kluwer Academic Publishers, Dordrecht.",
                "links": null
            },
            "BIBREF9": {
                "ref_id": "b9",
                "title": "Noun classification from predicateargument structures",
                "authors": [
                    {
                        "first": "D",
                        "middle": [],
                        "last": "Hindle",
                        "suffix": ""
                    }
                ],
                "year": 1990,
                "venue": "Meeting of the Association for Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "268--275",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "D. Hindle. 1990. Noun classification from predicate- argument structures. In Meeting of the Association for Computational Linguistics, pages 268-275.",
                "links": null
            },
            "BIBREF10": {
                "ref_id": "b10",
                "title": "Cluto: A clustering toolkit",
                "authors": [
                    {
                        "first": "G",
                        "middle": [],
                        "last": "Karypis",
                        "suffix": ""
                    }
                ],
                "year": 2003,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "G. Karypis. 2003. Cluto: A clustering toolkit. tech- nical report 02-017. Technical report, University of Minnesota, November.",
                "links": null
            },
            "BIBREF11": {
                "ref_id": "b11",
                "title": "Automatic retrieval and clustering of similar words",
                "authors": [
                    {
                        "first": "D",
                        "middle": [],
                        "last": "Lin",
                        "suffix": ""
                    }
                ],
                "year": 1998,
                "venue": "COLING-ACL",
                "volume": "",
                "issue": "",
                "pages": "768--774",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "D. Lin. 1998. Automatic retrieval and clustering of similar words. In COLING-ACL, pages 768-774.",
                "links": null
            },
            "BIBREF12": {
                "ref_id": "b12",
                "title": "Dependency-based construction of semantic space models",
                "authors": [
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Pad\u00f3",
                        "suffix": ""
                    },
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Lapata",
                        "suffix": ""
                    }
                ],
                "year": 2007,
                "venue": "Comput. Linguist",
                "volume": "33",
                "issue": "2",
                "pages": "161--199",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "S. Pad\u00f3 and M. Lapata. 2007. Dependency-based con- struction of semantic space models. Comput. Lin- guist., 33(2):161-199.",
                "links": null
            },
            "BIBREF13": {
                "ref_id": "b13",
                "title": "Identifying concept attributes using a classifier",
                "authors": [
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Poesio",
                        "suffix": ""
                    },
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Almuhareb",
                        "suffix": ""
                    }
                ],
                "year": 2005,
                "venue": "Proceedings of the ACL-SIGLEX Workshop on Deep Lexical Acquisition",
                "volume": "",
                "issue": "",
                "pages": "18--27",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "M. Poesio and A. Almuhareb. 2005b. Identifying con- cept attributes using a classifier. In Proceedings of the ACL-SIGLEX Workshop on Deep Lexical Acqui- sition, pages 18-27, Ann Arbor, Michigan, June. As- sociation for Computational Linguistics.",
                "links": null
            },
            "BIBREF14": {
                "ref_id": "b14",
                "title": "The Word Space Model",
                "authors": [
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Sahlgren",
                        "suffix": ""
                    }
                ],
                "year": 2006,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "M. Sahlgren. 2006. The Word Space Model. Ph.D. thesis, Department of Linguistics, Stockholm Uni- versity.",
                "links": null
            },
            "BIBREF15": {
                "ref_id": "b15",
                "title": "Dimensions of meaning",
                "authors": [
                    {
                        "first": "H",
                        "middle": [],
                        "last": "Sch\u00fctze",
                        "suffix": ""
                    }
                ],
                "year": 1992,
                "venue": "Supercomputing '92: Proceedings of the 1992 ACM/IEEE conference on Supercomputing",
                "volume": "",
                "issue": "",
                "pages": "787--796",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "H. Sch\u00fctze. 1992. Dimensions of meaning. In Super- computing '92: Proceedings of the 1992 ACM/IEEE conference on Supercomputing, pages 787-796, Los Alamitos, CA, USA. IEEE Computer Society Press.",
                "links": null
            }
        },
        "ref_entries": {
            "FIGREF0": {
                "text": "Context definitions used in the construction of our word spaces. All examples show contexts for the target human. Greyed out parts are just for illustrative purposes and have no impact on the word vectors. The examples are slightly simplified versions of sentences found in ukWaC.",
                "num": null,
                "type_str": "figure",
                "uris": null
            },
            "TABREF0": {
                "content": "<table><tr><td/><td/><td/><td/><td/><td/><td/><td/><td colspan=\"4\">(II) an object of a verb</td><td/><td/></tr><tr><td/><td/><td colspan=\"2\">s u b j</td><td/><td/><td/><td/><td/><td/><td colspan=\"2\">su bj</td><td/><td>o b j</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td>V</td><td/></tr><tr><td/><td>p re</td><td/><td>V</td><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td>N</td></tr><tr><td colspan=\"2\">PreDet</td><td>N</td><td/><td/><td/><td/><td/><td colspan=\"3\">Gods from another world</td><td/><td colspan=\"2\">created</td><td>humans</td></tr><tr><td>All</td><td/><td>humans</td><td>die.</td><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td colspan=\"5\">(III) modified by an adjective</td><td/><td/><td/><td colspan=\"7\">(IV) linked to another noun via a genitive relation</td></tr><tr><td/><td>s</td><td>p re d</td><td/><td/><td/><td/><td/><td/><td/><td/><td/><td colspan=\"2\">s u b j</td><td>m od</td></tr><tr><td/><td/><td/><td/><td colspan=\"2\">pc om p-n</td><td/><td/><td/><td/><td colspan=\"2\">g e n</td><td/><td/></tr><tr><td/><td colspan=\"2\">VBE</td><td>Prep</td><td/><td>m o d</td><td/><td/><td>d e t</td><td/><td/><td>N</td><td/><td>V</td></tr><tr><td/><td/><td/><td/><td/><td/><td>N</td><td/><td/><td/><td>N</td><td/><td/><td/></tr><tr><td/><td/><td/><td/><td/><td>A</td><td/><td/><td>Det</td><td/><td/><td/><td/><td/></tr><tr><td>Young dogs</td><td colspan=\"2\">are</td><td>like</td><td colspan=\"2\">young</td><td colspan=\"2\">humans</td><td>The</td><td colspan=\"2\">human's</td><td colspan=\"2\">eyes</td><td colspan=\"2\">glimmered</td><td>with comprehension</td></tr><tr><td colspan=\"5\">(V) part of a nominal complex</td><td/><td/><td/><td colspan=\"5\">(VI) part of a conjunction</td><td/></tr><tr><td/><td>de t</td><td>n n</td><td>s u b j</td><td>V</td><td>o b j</td><td/><td/><td colspan=\"2\">p u n c</td><td>co nj</td><td>s</td><td/><td colspan=\"2\">VBE</td><td>p r e d</td></tr><tr><td/><td/><td>N</td><td/><td/><td/><td/><td/><td>N</td><td/><td/><td/><td/><td/></tr><tr><td>Det</td><td>N</td><td/><td/><td/><td/><td/><td/><td/><td/><td>U</td><td colspan=\"2\">N</td><td/></tr><tr><td>The</td><td>human</td><td colspan=\"2\">body</td><td colspan=\"2\">presents</td><td colspan=\"2\">a problem.</td><td>Humans</td><td/><td>and</td><td colspan=\"2\">animals</td><td colspan=\"2\">are</td><td>equally fair game.</td></tr><tr><td colspan=\"6\">(VII) the subject of a predicate noun</td><td/><td/><td/><td/><td colspan=\"5\">(VIII) the subject of a predicate adjective</td></tr><tr><td>s</td><td/><td/><td>pre d</td><td/><td/><td/><td/><td/><td/><td/><td/><td>s</td><td colspan=\"2\">p re d</td></tr><tr><td>N</td><td>VBE</td><td>subj</td><td colspan=\"2\">d e t, m o d</td><td>N</td><td>re l</td><td/><td>C</td><td/><td colspan=\"2\">Humans N</td><td/><td>are VBE subj</td><td>fallible. A</td></tr><tr><td>Humans</td><td>are</td><td colspan=\"2\">the only</td><td colspan=\"2\">specie</td><td colspan=\"3\">that has sex for pleasure.</td><td/><td/><td/><td/><td/></tr><tr><td colspan=\"9\">(IX) the prepositional complement modifying a noun</td><td/><td colspan=\"5\">(X) the prepositional complement modifying a</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td colspan=\"5\">noun that is the subject of a predicate adjective</td></tr><tr><td/><td>s</td><td>a u x</td><td/><td>ob j</td><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td/><td/><td/><td/><td/><td>d e t</td><td>m o d</td><td/><td/><td/><td/><td/><td/><td>s</td><td>p re d</td></tr><tr><td>You N</td><td>must Aux</td><td colspan=\"2\">get into V</td><td>the Det</td><td colspan=\"2\">mind N</td><td>of Prep</td><td>humans. N p c o m p -n</td><td/><td>N</td><td>m o d</td><td>Prep</td><td>p c o m p -n</td><td>N</td><td>VBE</td><td>A</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td/><td/><td colspan=\"2\">The nature</td><td>of</td><td colspan=\"2\">humans</td><td>is</td><td>corrupt.</td></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td>s</td><td/><td/><td>p re d</td><td/><td/></tr><tr><td/><td/><td/><td/><td/><td colspan=\"2\">m o d</td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td/><td/><td/><td/><td/><td>N</td><td/><td colspan=\"2\">p c o m p -n</td><td/><td>VBE</td><td/><td>N</td><td/></tr><tr><td/><td/><td/><td/><td/><td/><td colspan=\"2\">Prep</td><td/><td/><td/><td/><td/><td/></tr><tr><td/><td/><td/><td/><td/><td/><td/><td/><td>N</td><td/><td/><td/><td/><td/></tr><tr><td/><td/><td/><td/><td colspan=\"2\">Chief diseases</td><td colspan=\"2\">of</td><td>humans</td><td/><td>are</td><td colspan=\"3\">infections.</td></tr><tr><td colspan=\"4\">(XII) relations I-They interrogated him</td><td/><td>about</td><td>the</td><td/><td>human</td><td/><td>body</td><td>and</td><td/><td/><td>reproduction.</td></tr></table>",
                "text": "the subject of a verb IV and VI-XI above but now with the target as part of a complex noun phrase as shown for a conjunction relation (VI) in the example",
                "num": null,
                "type_str": "table",
                "html": null
            },
            "TABREF2": {
                "content": "<table><tr><td>association</td><td colspan=\"3\">g-score t-score t-score +</td></tr><tr><td>measure</td><td/><td/><td/></tr><tr><td>dependency</td><td colspan=\"2\">77.1% 85.5%</td><td>96.7%</td></tr><tr><td>based space</td><td/><td/><td/></tr><tr><td>window based</td><td colspan=\"2\">84.1% 82.7%</td><td>89.3%</td></tr><tr><td>space</td><td/><td/><td/></tr><tr><td>pattern based</td><td>-</td><td>-</td><td>85.5%</td></tr><tr><td>space</td><td/><td/><td/></tr></table>",
                "text": ".",
                "num": null,
                "type_str": "table",
                "html": null
            },
            "TABREF4": {
                "content": "<table><tr><td>: Clustering results using only one kind of</td></tr><tr><td>path specification. For (IV), (X) and (XI) purity</td></tr><tr><td>values are missing because vectors for some of the</td></tr><tr><td>words could not be built.</td></tr></table>",
                "text": "",
                "num": null,
                "type_str": "table",
                "html": null
            }
        }
    }
}