-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathcs224n_lec4.html
977 lines (660 loc) · 38.8 KB
/
cs224n_lec4.html
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
<!DOCTYPE html>
<html>
<head>
<!-- Document Settings -->
<meta charset="utf-8" />
<meta http-equiv="X-UA-Compatible" content="IE=edge" />
<!-- On Post front-matter YAML, set "use_math: true" to use LaTex -->
<script type="text/x-mathjax-config">
MathJax.Hub.Config({
TeX: {
equationNumbers: {
autoNumber: "AMS"
}
},
tex2jax: {
inlineMath: [ ['$', '$'], ["\\(","\\)"] ],
displayMath: [ ['$$', '$$'], ["\\[","\\]"] ],
processEscapes: true,
}
});
MathJax.Hub.Register.MessageHook("Math Processing Error",function (message) {
alert("Math Processing Error: "+message[1]);
});
MathJax.Hub.Register.MessageHook("TeX Jax - parse error",function (message) {
alert("Math Processing Error: "+message[1]);
});
</script>
<script type="text/javascript" async
src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.2/MathJax.js?config=TeX-MML-AM_CHTML">
</script>
<!-- Base Meta -->
<!-- dynamically fixing the title for tag/author pages -->
<title>cs224n - Lecture 4. Dependency Parsing</title>
<meta name="HandheldFriendly" content="True" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<!-- Styles'n'Scripts -->
<link rel="stylesheet" type="text/css" href="/assets/built/screen.css" />
<link rel="stylesheet" type="text/css" href="/assets/built/screen.edited.css" />
<link rel="stylesheet" type="text/css" href="/assets/built/syntax.css" />
<!-- syntax.css -->
<link rel="stylesheet" type="text/css" href="/assets/built/syntax.css" />
<!-- highlight.js -->
<link rel="stylesheet" href="//cdnjs.cloudflare.com/ajax/libs/highlight.js/9.12.0/styles/default.min.css">
<style>.hljs { background: none; }</style>
<!--[if IE]>
<style>
p, ol, ul{
width: 100%;
}
blockquote{
width: 100%;
}
</style>
<![endif]-->
<!-- This tag outputs SEO meta+structured data and other important settings -->
<meta name="description" content="" />
<link rel="shortcut icon" href="http://0.0.0.0:4000/assets/built/images/favicon.jpg" type="image/png" />
<link rel="canonical" href="http://0.0.0.0:4000/cs224n_lec4" />
<meta name="referrer" content="no-referrer-when-downgrade" />
<!--title below is coming from _includes/dynamic_title-->
<meta property="og:site_name" content="Darron's Devlog" />
<meta property="og:type" content="website" />
<meta property="og:title" content="cs224n - Lecture 4. Dependency Parsing" />
<meta property="og:description" content="Two views of linguistic structure: Phrase structure Constituency = phrase structure grammar = context-free grammers(CFGs) Phrase structure organizes words into nested constituents Starting unit: words (noun, preposition, adjective, determiner, …) the, $\ $ cat, $\ $ cuddly, $\ $ by, $\ $ door Words combine into phrases the cuddly cat(noun" />
<meta property="og:url" content="http://0.0.0.0:4000/cs224n_lec4" />
<meta property="og:image" content="http://0.0.0.0:4000/assets/built/images/blog-cover1.png" />
<meta property="article:publisher" content="https://www.facebook.com/" />
<meta property="article:author" content="https://www.facebook.com/" />
<meta property="article:published_time" content="2022-03-11T00:00:00+00:00" />
<meta property="article:modified_time" content="2022-03-11T00:00:00+00:00" />
<meta property="article:tag" content="cs224n" />
<meta name="twitter:card" content="summary_large_image" />
<meta name="twitter:title" content="cs224n - Lecture 4. Dependency Parsing" />
<meta name="twitter:description" content="Two views of linguistic structure: Phrase structure Constituency = phrase structure grammar = context-free grammers(CFGs) Phrase structure organizes words into nested constituents Starting unit: words (noun, preposition, adjective, determiner, …) the, $\ $ cat, $\ $ cuddly, $\ $ by, $\ $ door Words combine into phrases the cuddly cat(noun" />
<meta name="twitter:url" content="http://0.0.0.0:4000/" />
<meta name="twitter:image" content="http://0.0.0.0:4000/assets/built/images/blog-cover1.png" />
<meta name="twitter:label1" content="Written by" />
<meta name="twitter:data1" content="Darron's Devlog" />
<meta name="twitter:label2" content="Filed under" />
<meta name="twitter:data2" content="cs224n" />
<meta name="twitter:site" content="@" />
<meta name="twitter:creator" content="@" />
<meta property="og:image:width" content="1400" />
<meta property="og:image:height" content="933" />
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "Website",
"publisher": {
"@type": "Organization",
"name": "Darron's Devlog",
"logo": "http://0.0.0.0:4000/"
},
"url": "http://0.0.0.0:4000/cs224n_lec4",
"image": {
"@type": "ImageObject",
"url": "http://0.0.0.0:4000/assets/built/images/blog-cover1.png",
"width": 2000,
"height": 666
},
"mainEntityOfPage": {
"@type": "WebPage",
"@id": "http://0.0.0.0:4000/cs224n_lec4"
},
"description": "Two views of linguistic structure: Phrase structure Constituency = phrase structure grammar = context-free grammers(CFGs) Phrase structure organizes words into nested constituents Starting unit: words (noun, preposition, adjective, determiner, …) the, $\ $ cat, $\ $ cuddly, $\ $ by, $\ $ door Words combine into phrases the cuddly cat(noun"
}
</script>
<!-- <script type="text/javascript" src="https://demo.ghost.io/public/ghost-sdk.min.js?v=724281a32e"></script>
<script type="text/javascript">
ghost.init({
clientId: "ghost-frontend",
clientSecret: "f84a07a72b17"
});
</script> -->
<meta name="generator" content="Jekyll 3.6.2" />
<link rel="alternate" type="application/rss+xml" title="cs224n - Lecture 4. Dependency Parsing" href="/feed.xml" />
</head>
<body class="post-template">
<div class="site-wrapper">
<!-- All the main content gets inserted here, index.hbs, post.hbs, etc -->
<!-- default -->
<!-- The tag above means: insert everything in this file
into the {body} of the default.hbs template -->
<header class="site-header outer">
<div class="inner">
<nav class="site-nav">
<div class="site-nav-left">
<a class="site-nav-logo" href="/">Darron's Devlog</a>
<ul class="nav" role="menu">
<li class="nav-home" role="menuitem"><a href="/">Home</a></li>
<li class="nav-about" role="menuitem"><a href="/about/">About</a></li>
<li class="nav-projects" role="menuitem"><a href="/tag/projects/">Projects</a></li>
<li class="nav-studies" role="menuitem"><a href="/tag/studies/">Studies</a></li>
<li class="nav-blog" role="menuitem"><a href="/tag/blog/">Blog</a></li>
<li class="nav-archive" role="menuitem">
<a href="/archive.html">All Posts</a>
</li>
</ul>
</div>
<div class="site-nav-right">
<div class="social-links">
</div>
<a class="subscribe-button" href="#subscribe">Search</a>
</div>
</nav>
</div>
</header>
<!-- Everything inside the #post tags pulls data from the post -->
<!-- #post -->
<main id="site-main" class="site-main outer" role="main">
<div class="inner">
<article class="post-full tag-cs224n no-image">
<header class="post-full-header">
<section class="post-full-meta">
<time class="post-full-meta-date" datetime="11 March 2022">11 March 2022</time>
<span class="date-divider">/</span>
<a href='/tag/cs224n/'>CS224N</a>
</section>
<h1 class="post-full-title">cs224n - Lecture 4. Dependency Parsing</h1>
</header>
<!--
-->
<section class="post-full-content">
<div class="kg-card-markdown">
<h3 id="two-views-of-linguistic-structure-phrase-structure">Two views of linguistic structure: Phrase structure</h3>
<ul>
<li>
<p>Constituency = phrase structure grammar = context-free grammers(CFGs)<br />
<strong>Phrase structure</strong> organizes words into nested constituents</p>
</li>
<li>
<p>Starting unit: <strong>words</strong> (noun, preposition, adjective, determiner, …)<br />
the, $\ $ cat, $\ $ cuddly, $\ $ by, $\ $ door</p>
</li>
<li>
<p>Words combine into <strong>phrases</strong><br />
the cuddly cat(noun phrase),<br />
by the door(prepositional phrase; preposition(by) + noun phrase)</p>
</li>
<li>
<p>Phrases can combine into bigger phrases<br />
the cuddly cat by the door(noun phrase)</p>
</li>
<li>Lexicon:<br />
$\text{N} \rightarrow \text{cat}$<br />
$\text{N} \rightarrow \text{door}$<br />
$\text{Det} \rightarrow \text{the}$<br />
$P \rightarrow \text{by}$<br />
$\text{Adj} \rightarrow \text{cuddly}$<br />
$\vdots$</li>
<li>Grammar:<br />
$\text{NP} \rightarrow \text{Det } \ \text{ (Adj)}^{\ast} \ \text{ N } \ \text{ (PP)}$<br />
$\text{PP} \rightarrow \text{P } \ \text{ NP}$<br />
$\vdots$</li>
</ul>
<h3 id="two-views-of-linguistic-structure-dependency-structure">Two views of linguistic structure: Dependency structure</h3>
<ul>
<li>Dependency structure shows which words depend on (modify, attach to, or are arguments of) which other words.</li>
</ul>
<h3 id="why-do-we-need-sentence-structure">Why do we need sentence structure?</h3>
<ul>
<li>Humans communicate complex ideas by composing words together into bigger units to convey complex meanings</li>
<li>Listeners need to work out what modifies <em>attaches to</em> what</li>
<li>A model needs to understand sentence structure in order to be able to interpret language correctly</li>
</ul>
<h3 id="ambiguities">Ambiguities</h3>
<ul>
<li>Prepositional phrase ambiguity
<ul>
<li>A key parsing decision is how we ‘attach’ various constituents
<ul>
<li><em>PP</em>s, adverbial or participial phrases, infinitives, coordinations<br />
e.g.<br />
\(\begin{align*}
\text{The board approved [its acquisition]} & \text{[by Royal Trustco Ltd.]} \\
& \text{[of Toronto]} \\
& \text{[for \$27 a share]} \\
& \text{[at its monthly meeting].}
\end{align*}\)</li>
</ul>
</li>
<li>With a sentence of <em>k</em> prepositional phrases at the end of it, the number of parses is given by the Catalan numbers; $C_n = (2n)!/[(n+1)!n!]$, an exponential series growing as the number of prepositional phrases.</li>
</ul>
</li>
<li>
<p>Coordination scopre ambiguity<br />
e.g. <strong>Shuttle veteran</strong> and longtime NASA executive__ Fred Gregory appointed to board<br />
$\rightarrow$ 1 or 2 person?</p>
</li>
<li>
<p>Adjectival/Adverbial Modifier ambiguity<br />
e.g. Students get <strong>first hand job experience</strong></p>
</li>
<li>Verb Phrase(VP) attachment ambiguity<br />
e.g. Mutilated body washes up on Rio beach <strong>to be used for Olympics beach volleyball</strong></li>
</ul>
<h3 id="dependency-paths-help-extract-semantic-interpretation">Dependency paths help extract semantic interpretation</h3>
<ul>
<li>simple practical example: extracting protein-protein interaction</li>
</ul>
<p><img src="/assets/images/cs224n/lec4_0.png" alt="png" width="80%", height="100%" /></p>
<h3 id="dependency-grammar-and-dependency-structure">Dependency Grammar and Dependency Structure</h3>
<ul>
<li>Dependency syntax postulates that syntactic structure consists of relations between lexical items, normally binary asymmetric relations (“arrows”) called <strong>dependencies</strong></li>
</ul>
<p><img src="/assets/images/cs224n/lec4_1.png" alt="png" width="60%", height="100%" /></p>
<ul>
<li>
<p>The arrows are commonly <strong>typed</strong> with the name of grammatical relations (subject, prepositional object, apposition, etc.)</p>
</li>
<li>An arrow connects a <strong>head</strong>(governor, superior, regent) with a <strong>dependent</strong>(modifier, inferior, subordinate)</li>
<li>
<p>Usually, dependencies form a tree(a connected, acyclic, single-root graph)</p>
</li>
<li>
<p>Check: some people draw the arrows one way; some the other way</p>
</li>
<li>Usually add a fake ROOT so every word is a dependent of precisely 1 other node</li>
</ul>
<h3 id="the-rise-of-annotated-data--universal-dependencies-treebanks">The rise of annotated data & Universal Dependencies treebanks</h3>
<ul>
<li>Advantages of treebank
<ul>
<li>Reusability of the labor
<ul>
<li>Many parsers, part-of-speech taggers, etc. can be built on it</li>
<li>Valuable resource for linguistic</li>
</ul>
</li>
<li>Broad coverage, not just a few intuitions</li>
<li>Frequencies and distributional information(statistics)</li>
<li>A way to evaluate NLP systems</li>
</ul>
</li>
</ul>
<h3 id="dependency-conditioning-preferences">Dependency Conditioning Preferences</h3>
<p><img src="/assets/images/cs224n/lec4_2.png" alt="png" width="60%", height="100%" /></p>
<ul>
<li>The sources of information for dependency parsing
<ol>
<li>Bilexical affinities: The dependency $\text{discussion}\rightarrow\text{issues}$ is plausible</li>
<li>Dependency distance: Most dependencies are between nearby words</li>
<li>Intervening material: Dependencies rarely span intervening verbs or punctuation</li>
<li>Valency of heads: How many dependents on which side are usual for a head?</li>
</ol>
</li>
</ul>
<h3 id="dependency-parsing">Dependency Parsing</h3>
<ul>
<li>A sentence is parsed by choosing for each word what other word (including ROOT) it is a dependent of</li>
<li>Usually some constraints:
<ul>
<li>Only one word is a dependent of ROOT</li>
<li>Don’t want cycles $A\rightarrow B$, $B\rightarrow A$</li>
</ul>
</li>
<li>This makes the dependencies a tree</li>
<li>Final issue is whether arrows can cross(be <strong>non-projective</strong>) or not</li>
</ul>
<h3 id="projectivity">Projectivity</h3>
<ul>
<li>Definition of a <strong>projective parse</strong>: There are no crossing dependency arcs when the words are laid out in their linear order, with all arcs above the words</li>
<li>Dependencies corresponding to a CFG tree must be <strong>projective</strong>
<ul>
<li>i.e., by forming dependencies by taking 1 child of each category as head</li>
</ul>
</li>
<li>Most syntactic structure is projective like this, but dependency theory normally does allow non-projective structures to account for displaced constituents
<ul>
<li>You can’t easily get the semantics of certain constructions right without these nonprojective dependencies</li>
</ul>
</li>
<li>e.g.<br />
<strong>From</strong> who did Bill buy the coffee yesterday?<br />
Who did Bill buy the coffee <strong>from</strong> yesterday</li>
</ul>
<h3 id="methods-of-dependency-parsing">Methods of Dependency Parsing</h3>
<ol>
<li>Dynamic programming, <em>Eisner(1996)</em>: $O(n^3)$ complexity</li>
<li>Graph algorithms, <em>McDonald et al.(2005)</em>: creating a Minimun Spanning Tree</li>
<li>Constraint Satisfaction, <em>Karlsson(1990)</em></li>
<li>“Transition-based parsing” or “deterministic dependency parsing”<br />
Greedy choice of attachments guided by good machine learning classifiers<br />
E.g., MaltParser, <em>Nivre et al.(2008)</em></li>
</ol>
<h3 id="greedy-transition-based-parsing-nivre-2003">Greedy transition-based parsing, Nivre 2003</h3>
<ul>
<li>A simple form of greedy discriminative dependency parser</li>
<li>The parser does a sequence of bottom-up actions
<ul>
<li>Roughly like “shift” or “reduce” in a shift-reduce parser, but the “reduce” actions are specialized to create dependencies with head on left or right</li>
</ul>
</li>
<li>The parser has:
<ul>
<li>a stack $\sigma$, written with top to the right
<ul>
<li>which starts with the ROOT symbol</li>
</ul>
</li>
<li>a buffer $\beta$, written with top to the left
<ul>
<li>which starts with the input sentence</li>
</ul>
</li>
<li>a set of dependency arcs A
<ul>
<li>which starts off empty</li>
</ul>
</li>
<li>a set of actions</li>
</ul>
</li>
</ul>
<h3 id="basic-transition-based-dependency-parser">Basic transition-based dependency parser</h3>
<p><img src="/assets/images/cs224n/lec4_3.png" alt="png" width="80%", height="100%" /></p>
<ul>
<li>Arc-standard transition-based parser<br />
E.g., Analysis of “I ate fish”</li>
</ul>
<p><img src="/assets/images/cs224n/lec4_4.png" alt="png" width="50%", height="100%" /><br />
<img src="/assets/images/cs224n/lec4_5.png" alt="png" width="80%", height="100%" /></p>
<h3 id="maltparser-nivre-and-hall-2005">MaltParser, Nivre and Hall 2005</h3>
<ul>
<li>How we choose the next action?<br />
Answer: Machine Learning!</li>
<li>Each action is predicted by a discriminative classifier (e.g., softmax classifier) over legal move
<ul>
<li>Max of 3 untyped choices; max of $\lvert R \rvert \times 2 + 1 $ when typed</li>
<li>Features: top of stack word, POS; first in buffer word, POS; etc.</li>
</ul>
</li>
<li>There is NO search(in the simplest form)
<ul>
<li>But you can profitably do a beam search if you wish(slower but better): You keep <em>k</em> good parse prefixes at each time step</li>
</ul>
</li>
<li>The model’s accuracy is <em>fractionally</em> below the state of the art in dependency parsing, but it provides <strong>very fast linear time parsing</strong>, with high accuracy, great for parsing the web</li>
</ul>
<h3 id="conventional-feature-representation">Conventional Feature Representation</h3>
<p><img src="/assets/images/cs224n/lec4_6.png" alt="png" width="80%", height="100%" /></p>
<h3 id="evaluation-of-dependency-parsing-labeled-dependency-accuracy">Evaluation of Dependency Parsing: (labeled) dependency accuracy</h3>
<p><img src="/assets/images/cs224n/lec4_7.png" alt="png" width="80%", height="100%" /></p>
</div>
</section>
<!-- Email subscribe form at the bottom of the page -->
<!--
<section class="subscribe-form">
<h3 class="subscribe-form-title">Subscribe to Darron's Devlog</h3>
<p>Get the latest posts delivered right to your inbox</p>
<span id="searchform" method="post" action="/subscribe/" class="">
<input class="confirm" type="hidden" name="confirm" />
<input class="location" type="hidden" name="location" />
<input class="referrer" type="hidden" name="referrer" />
<div class="form-group">
<input class="subscribe-email" onkeyup="myFunc()"
id="searchtext" type="text" name="searchtext"
placeholder="Search..." />
</div>
<script type="text/javascript">
function myFunc() {
if(event.keyCode == 13) {
var url = encodeURIComponent($("#searchtext").val());
location.href = "/search.html?query=" + url;
}
}
</script>
</span>
</section>
-->
<footer class="post-full-footer">
<!-- Everything inside the #author tags pulls data from the author -->
<!-- #author-->
<!-- /author -->
</footer>
<!-- If you use Disqus comments, just uncomment this block.
The only thing you need to change is "test-apkdzgmqhj" - which
should be replaced with your own Disqus site-id. -->
<section class="post-full-comments">
<div id="disqus_thread"></div>
<script>
var disqus_config = function () {
var this_page_url = 'http://0.0.0.0:4000/cs224n_lec4';
var this_page_identifier = '/cs224n_lec4';
var this_page_title = 'cs224n - Lecture 4. Dependency Parsing';
};
(function() {
var d = document, s = d.createElement('script');
s.src = 'https://.disqus.com/embed.js';
s.setAttribute('data-timestamp', +new Date());
(d.head || d.body).appendChild(s);
})();
</script>
</section>
</article>
</div>
</main>
<!-- Links to Previous/Next posts -->
<aside class="read-next outer">
<div class="inner">
<div class="read-next-feed">
<article class="read-next-card"
style="background-image: url(/assets/built/images/blog-cover1.png)"
>
<header class="read-next-card-header">
<small class="read-next-card-header-sitetitle">— Darron's Devlog —</small>
<h3 class="read-next-card-header-title"><a href="/tag/cs224n/">Cs224n</a></h3>
</header>
<div class="read-next-divider"><svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M13 14.5s2 3 5 3 5.5-2.463 5.5-5.5S21 6.5 18 6.5c-5 0-7 11-12 11C2.962 17.5.5 15.037.5 12S3 6.5 6 6.5s4.5 3.5 4.5 3.5"/></svg>
</div>
<div class="read-next-card-content">
<ul>
<li><a href="/cs224n_lec10">cs224n - Lecture 10. Transformers and Pretraining</a></li>
<li><a href="/cs224n_lec9">cs224n - Lecture 9. Self-Attention and Transformers</a></li>
<li><a href="/cs224n_lec8">cs224n - Lecture 8. Attention (Contd.)</a></li>
</ul>
</div>
<footer class="read-next-card-footer">
<a href="/tag/cs224n/">
See all 9 posts →
</a>
</footer>
</article>
<!-- If there's a next post, display it using the same markup included from - partials/post-card.hbs -->
<article class="post-card post-template no-image">
<div class="post-card-content">
<a class="post-card-content-link" href="/cs224n_lec5">
<header class="post-card-header">
<span class="post-card-tags">Cs224n</span>
<h2 class="post-card-title">cs224n - Lecture 5. Language Models and RNNs</h2>
</header>
<section class="post-card-excerpt">
<p>How do we gain from a neural dependency parser? So far… Transition based dependency parsers were an efficient linear time method for giving the syntactic structure of natural language text. Worked pretty well</p>
</section>
</a>
<footer class="post-card-meta">
</footer>
</div>
</article>
<!-- If there's a previous post, display it using the same markup included from - partials/post-card.hbs -->
<article class="post-card post-template no-image">
<div class="post-card-content">
<a class="post-card-content-link" href="/cs224n_lec3">
<header class="post-card-header">
<span class="post-card-tags">Cs224n</span>
<h2 class="post-card-title">cs224n - Lecture 3. Backprop and Neural Networks</h2>
</header>
<section class="post-card-excerpt">
<p>Named Entity Recognition(NER) Task: find and classify names in text Possible uses: Tracking mentions of particular entities in documents For question answering, answers are usually named entities Often followed by Named Entity Linking/Canonicalization</p>
</section>
</a>
<footer class="post-card-meta">
</footer>
</div>
</article>
</div>
</div>
</aside>
<!-- Floating header which appears on-scroll, included from includes/floating-header.hbs -->
<div class="floating-header">
<div class="floating-header-logo">
<a href="/">
<span>Darron's Devlog</span>
</a>
</div>
<span class="floating-header-divider">—</span>
<div class="floating-header-title">cs224n - Lecture 4. Dependency Parsing</div>
<div class="floating-header-share">
<div class="floating-header-share-label">Share this <svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24">
<path d="M7.5 15.5V4a1.5 1.5 0 1 1 3 0v4.5h2a1 1 0 0 1 1 1h2a1 1 0 0 1 1 1H18a1.5 1.5 0 0 1 1.5 1.5v3.099c0 .929-.13 1.854-.385 2.748L17.5 23.5h-9c-1.5-2-5.417-8.673-5.417-8.673a1.2 1.2 0 0 1 1.76-1.605L7.5 15.5zm6-6v2m-3-3.5v3.5m6-1v2"/>
</svg>
</div>
<a class="floating-header-share-tw" href="https://twitter.com/share?text=cs224n+-+Lecture+4.+Dependency+Parsing&url=https://12kdh43.github.io/cs224n_lec4"
onclick="window.open(this.href, 'share-twitter', 'width=550,height=235');return false;">
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 32 32"><path d="M30.063 7.313c-.813 1.125-1.75 2.125-2.875 2.938v.75c0 1.563-.188 3.125-.688 4.625a15.088 15.088 0 0 1-2.063 4.438c-.875 1.438-2 2.688-3.25 3.813a15.015 15.015 0 0 1-4.625 2.563c-1.813.688-3.75 1-5.75 1-3.25 0-6.188-.875-8.875-2.625.438.063.875.125 1.375.125 2.688 0 5.063-.875 7.188-2.5-1.25 0-2.375-.375-3.375-1.125s-1.688-1.688-2.063-2.875c.438.063.813.125 1.125.125.5 0 1-.063 1.5-.25-1.313-.25-2.438-.938-3.313-1.938a5.673 5.673 0 0 1-1.313-3.688v-.063c.813.438 1.688.688 2.625.688a5.228 5.228 0 0 1-1.875-2c-.5-.875-.688-1.813-.688-2.75 0-1.063.25-2.063.75-2.938 1.438 1.75 3.188 3.188 5.25 4.25s4.313 1.688 6.688 1.813a5.579 5.579 0 0 1 1.5-5.438c1.125-1.125 2.5-1.688 4.125-1.688s3.063.625 4.188 1.813a11.48 11.48 0 0 0 3.688-1.375c-.438 1.375-1.313 2.438-2.563 3.188 1.125-.125 2.188-.438 3.313-.875z"/></svg>
</a>
<a class="floating-header-share-fb" href="https://www.facebook.com/sharer/sharer.php?u=https://12kdh43.github.io/cs224n_lec4"
onclick="window.open(this.href, 'share-facebook','width=580,height=296');return false;">
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 32 32"><path d="M19 6h5V0h-5c-3.86 0-7 3.14-7 7v3H8v6h4v16h6V16h5l1-6h-6V7c0-.542.458-1 1-1z"/></svg>
</a>
</div>
<progress class="progress" value="0">
<div class="progress-container">
<span class="progress-bar"></span>
</div>
</progress>
</div>
<!-- /post -->
<!-- The #contentFor helper here will send everything inside it up to the matching #block helper found in default.hbs -->
<!-- Previous/next page links - displayed on every page -->
<!-- The footer at the very bottom of the screen -->
<footer class="site-footer outer">
<div class="site-footer-content inner">
<section class="copyright"><a href="/">Darron's Devlog</a> © 2022</section>
<!--
<section class="poweredby">Proudly published with <a href="https://jekyllrb.com/">Jekyll</a> &
<a href="https://pages.github.com/" target="_blank" rel="noopener">GitHub Pages</a> using
<a href="https://github.com/jekyllt/jasper2" target="_blank" rel="noopener">Jasper2</a></section>
-->
<nav class="site-footer-nav">
<a href="/">Latest Posts</a>
</nav>
</div>
</footer>
</div>
<!-- The big email subscribe modal content -->
<div id="subscribe" class="subscribe-overlay">
<a class="subscribe-overlay-close" href="#"></a>
<div class="subscribe-overlay-content">
<h1 class="subscribe-overlay-title">Search Darron's Devlog</h1>
<p class="subscribe-overlay-description">
</p>
<span id="searchform" method="post" action="/subscribe/" class="">
<input class="confirm" type="hidden" name="confirm" />
<input class="location" type="hidden" name="location" />
<input class="referrer" type="hidden" name="referrer" />
<div class="form-group">
<input class="subscribe-email" onkeyup="myFunc()"
id="searchtext" type="text" name="searchtext"
placeholder="Search..." />
</div>
<script type="text/javascript">
function myFunc() {
if(event.keyCode == 13) {
var url = encodeURIComponent($("#searchtext").val());
location.href = "/search.html?query=" + url;
}
}
</script>
</span>
</div>
</div>
<!-- highlight.js -->
<script src="https://cdnjs.cloudflare.com/ajax/libs/prism/1.10.0/components/prism-abap.min.js"></script>
<script>$(document).ready(function() {
$('pre code').each(function(i, block) {
hljs.highlightBlock(block);
});
});</script>
<!-- jQuery + Fitvids, which makes all video embeds responsive -->
<script
src="https://code.jquery.com/jquery-3.2.1.min.js"
integrity="sha256-hwg4gsxgFZhOsEEamdOYGBf13FyQuiTwlAQgxVSNgt4="
crossorigin="anonymous">
</script>
<script type="text/javascript" src="/assets/js/jquery.fitvids.js"></script>
<script type="text/javascript" src="https://demo.ghost.io/assets/js/jquery.fitvids.js?v=724281a32e"></script>
<!-- Paginator increased to "infinit" in _config.yml -->
<!-- if paginator.posts -->
<!-- <script>
var maxPages = parseInt('');
</script>
<script src="/assets/js/infinitescroll.js"></script> -->
<!-- /endif -->
<!-- Add Google Analytics -->
<!-- Google Analytics Tracking code -->
<script>
(function(i,s,o,g,r,a,m){i['GoogleAnalyticsObject']=r;i[r]=i[r]||function(){
(i[r].q=i[r].q||[]).push(arguments)},i[r].l=1*new Date();a=s.createElement(o),
m=s.getElementsByTagName(o)[0];a.async=1;a.src=g;m.parentNode.insertBefore(a,m)
})(window,document,'script','//www.google-analytics.com/analytics.js','ga');
ga('create', '', 'auto');
ga('send', 'pageview');
</script>
<!-- The #block helper will pull in data from the #contentFor other template files. In this case, there's some JavaScript which we only want to use in post.hbs, but it needs to be included down here, after jQuery has already loaded. -->
<script>
// NOTE: Scroll performance is poor in Safari
// - this appears to be due to the events firing much more slowly in Safari.
// Dropping the scroll event and using only a raf loop results in smoother
// scrolling but continuous processing even when not scrolling
$(document).ready(function () {
// Start fitVids
var $postContent = $(".post-full-content");
$postContent.fitVids();
// End fitVids
var progressBar = document.querySelector('progress');
var header = document.querySelector('.floating-header');
var title = document.querySelector('.post-full-title');
var lastScrollY = window.scrollY;
var lastWindowHeight = window.innerHeight;
var lastDocumentHeight = $(document).height();
var ticking = false;
function onScroll() {
lastScrollY = window.scrollY;
requestTick();
}
function onResize() {
lastWindowHeight = window.innerHeight;
lastDocumentHeight = $(document).height();
requestTick();
}
function requestTick() {
if (!ticking) {
requestAnimationFrame(update);
}
ticking = true;
}
function update() {
var trigger = title.getBoundingClientRect().top + window.scrollY;
var triggerOffset = title.offsetHeight + 35;
var progressMax = lastDocumentHeight - lastWindowHeight;
// show/hide floating header
if (lastScrollY >= trigger + triggerOffset) {
header.classList.add('floating-active');
} else {
header.classList.remove('floating-active');
}
progressBar.setAttribute('max', progressMax);
progressBar.setAttribute('value', lastScrollY);
ticking = false;
}
window.addEventListener('scroll', onScroll, {passive: true});
window.addEventListener('resize', onResize, false);
update();
});
</script>
<!-- Ghost outputs important scripts and data with this tag - it should always be the very last thing before the closing body tag -->
<!-- ghost_foot -->
</body>
</html>