ie_relation_extraction.tex 17.5 KB
Newer Older
Loïc Barrault's avatar
ie  
Loïc Barrault committed
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
% !TEX root = text_processing.tex
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{}

\vfill
\centering
\Huge{\edinred{[Information Extraction]\\Relation Extraction}}

\end{frame}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Information Extraction: Overview}

\begin{itemize}
\item \gray{Introduction to Information Extraction}
\begin{itemize}
	\item \gray{Definition + contrast with IR}
	\item \gray{Example Applications}
	\item \gray{Overview of Tasks}
	\item \gray{Overview of Approaches}
	\item \gray{Evaluation + Shared Task Challenges}
	\item \gray{Brief(est) history of IE}
\end{itemize}

\item \gray{Named Entity Recognition}
\begin{itemize}
	\item \gray{Task}
	\item \gray{Approaches: Rule-based, Supervised Learning}
	\item \gray{Entity Linking}
\end{itemize}

\item \textbf{Relation Extraction}
\begin{itemize}
	\item \textbf{Task}
	\item \textbf{Approaches: Rule-based, Supervised Learning, Bootstrapping, Distant Supervision}
\end{itemize}
\end{itemize}

\end{frame}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Information Extraction: Relation Extraction: Overview}

\textbf{Relation Extraction}
\begin{itemize}
\item Task definition
\end{itemize}

\textbf{Approaches to Relation Extraction}
\begin{itemize}
\item Knowledge-engineering approaches to RE
\item Supervised learning approaches to RE
\item Bootstrapping Approaches to RE
\item Distant Supervision Approaches to RE
\end{itemize}

\end{frame}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Information Extraction: Relation Extraction - recap}

\begin{block}{\textbf{Relation Extraction (RE)}}
Task: Identify all \myemph{assertions of relations} holding between \myemph{entities} in a text $T$\\
\begin{itemize}
\item the entities are identified in previous entity extraction step
\item The set of possible relations $\mathbf{R}$ is determined in advance
\end{itemize}
\end{block}

\textbf{Note:}
\begin{itemize}
\item relations in $\mathbf{R}$ are usually binary
\item the entity classes involved in a relation in $\mathbf{R}$ are assumed to be a subset of those identified in the entity extraction process
\end{itemize}

\textbf{Generally divided into two subtasks:}
\begin{enumerate}
\item \myemph{Relation detection}: find pairs of entities between which a relation holds
\item \myemph{Relation classification}: determine the type of a previously detected relation 
\end{enumerate}

\end{frame}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Information Extraction: Relation Extraction: Examples}

\textbf{Some examples}

\begin{itemize}
\item \textsc{location\_of} holding between:
\begin{itemize}
	\item \textsc{organisation} and \textsc{geopolitical\_location}
	\item medical \textsc{investigation} and \textsc{ body\_part}
	\item \textsc{gene} and \textsc{chromosome\_location}
\end{itemize}

\item \textsc{employee\_of} holding between \textsc{person} and \textsc{organisation}\\
\item \textsc{product\_of} holding between \textsc{artifact} and \textsc{organisation}\\
\item \textsc{is\_exposed\_to} holding between \textsc{organisation} and \textsc{risk}\\
\item \textsc{is\_associated\_with} holding between \textsc{drug} and \textsc{side\_effect}\\
\item \textsc{interaction} holding between \textsc{protein} and \textsc{protein}\\
\end{itemize}
\end{frame}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Information Extraction: RE: Challenges}

\textbf{Challenges for Relation Extraction}
\begin{enumerate}
\item<1-> Many different ways to express the same relation
\begin{itemize}
\item \myemph{Canonical}: \annot{Microsoft}{ORG} \textbf{\underline{is located in}} \annot{Redmond}{LOC}
\item \myemph{Synonyms}: \annot{Microsoft}{ORG} \textbf{\underline{is located/based/headquartered in}}  \annot{Redmond}{LOC}
\item \myemph{Syntactic variations}:
\begin{itemize}
	\item \annot{Microsoft}{ORG}, the software giant and ... , \textbf{\underline{is based in}} \annot{Redmond}{LOC}
	\item \annot{Redmond}{LOC}\textbf{\underline{-based}} \annot{Microsoft}{ORG} ...
	\item \annot{Redmond}{LOC}\textbf{\underline{'s}} \annot{Microsoft}{ORG} ...
	\item \annot{Redmond}{LOC} software giant \annot{Microsoft}{ORG} ...
\end{itemize}
\end{itemize}

\item<2-> Relations often involve coreference links\\
\small{
\myemph{\annot{Bill Gates}{PER}} \textbf{\underline{is co-founder, technology advisor and board member}} of \annot{Microsoft}{ORG}. \myemph{\annot{He}{PER}} \textbf{\underline{served as chairman of the board}} until Feb. 4, 2014. On June 27, 2008, \myemph{\annot{Gates}{PER}} \textbf{\underline{transitioned out}} of a day-to-day role in the company to spend more time on his global health and education work at the \annot{Bill \& Melinda Gates Foundation}{ORG}.
}
\end{enumerate}

\end{frame}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Information Extraction: RE: Challenges}

\textbf{Challenges for Relation Extraction}
\begin{enumerate}
\setcounter{enumi}{2}
\item The information required may be spread across multiple sentences
\item The information may be implied by the text rather than explicitly asserted \ra\ \myemphb{inference}
\end{enumerate}

\textbf{Ex1.:}
\myemph{\annot{Dirk Ruthless}{PER}} of \annot{MegaCorp}{ORG} made a stunning announcement today. In September \myemph{\annot{he}{PER}} will be stepping down as \annot{Chief Executive Officer}{POS} to spend more time with his pet piranhas.

\begin{itemize}
\item[\Ra] resolve \textbf{pronominal anaphor}  \myemph{Dirk Ruthless} \Lra\ \myemph{he} to determine the corporate \textsc{position}
\item[\Ra] In Ex1.: no explicit statement that \myemph{Dirk Ruthless} \textbf{is} CEO of MegaCorp
\begin{itemize}
	\item "\myemph{\annot{Dirk Ruthless}{PER}} of \annot{MegaCorp}{ORG}"  + "will be stepping down as \annot{Chief Executive Officer}{POS}" = \myemph{\annot{Dirk Ruthless}{PER}} is CEO of \annot{MegaCorp}{ORG}
\end{itemize}
\item Solving RE may imply solving \myemphb{textual entailment}
\end{itemize}

\end{frame}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Information Extraction: Relation Extraction: Overview}

\gray{Relation Extraction}
\begin{itemize}
\item \gray{Task definition}
\end{itemize}

\textbf{Approaches to Relation Extraction}
\begin{itemize}
\item \textbf{Knowledge-engineering approaches to RE}
\item Supervised learning approaches to RE
\item Bootstrapping Approaches to RE
\item Distant Supervision Approaches to RE
\end{itemize}

\end{frame}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Information Extraction: RE: Knowledge-engineering}

\colorbox{liumlightgray}{
\parbox{.99\textwidth}{

\annot{Mr. \tikzmark{a} Wright}{PER}, \textbf{\underline{\annot{executive vice president}{POSITION}}} of \annot{Merrill Lynch \tikzmark{b} Canada Inc.}{ORG}\\

\begin{tikzpicture}[overlay,remember picture]
 \draw [very thick, color=carminered] ($({pic cs:a})+(0ex,-1ex)$) -- ($({pic cs:a})+(0ex,-3ex)$);
 \draw [very thick, color=carminered] ($({pic cs:a})+(0ex,-3ex)$) -- ($({pic cs:b})+(0ex,-3ex)$)  node [midway, below, color=carminered] {is-employed-by};
 \draw [very thick, color=carminered] ($({pic cs:b})+(0ex,-3ex)$) -- ($({pic cs:b})+(0ex,-1ex)$);
\end{tikzpicture}

}}

\vfill

Such systems use manually authored rules and can be divided into
\begin{itemize}
\item<1-> “\myemph{shallow}”: systems engineered to the IE task, typically using \myemphb{pattern-action} rules
\begin{center}
\begin{tabular}{ll}
Pattern: & \textbf{"\$Person, \$Position of \$Organization"} \\
Action: & \textbf{add-relation(is-employed-by(\$Person,\$Organization))} \\
\end{tabular}
\end{center}
\vspace{.5cm}
\item<2-> “\myemph{deep}”: linguistically inspired language understanding systems
\end{itemize}

\vfill

\end{frame}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Information Extraction: RE: Knowledge-engineering}

\begin{itemize}
\item\myemph{deep}”: linguistically inspired language understanding systems
\begin{itemize}
	\item parse input to identify key grammatical relations (e.g. subject, object)
	\item use transduction rules on parser output to extract relations 
	\item[\ra] more powerful than regex on NE tags alone
\end{itemize}
\end{itemize}
\vspace{3.5cm}

\only<2->{
\begin{itemize}
\item Ex. parse trees:
\begin{itemize}
\item Multiple surface forms share underlying syntactic structure
\item \textsc{subject} = PER, \textsc{object} = ORG and \textsc{verb} = \textbf{works for}
\end{itemize}
\end{itemize}

\begin{textblock*}{50mm}[0,0](115mm,25mm)
\includegraphics[height=.45\textheight]{syntaxtree1}
\end{textblock*}

\begin{textblock*}{50mm}[0,0](55mm,25mm)
\includegraphics[height=.45\textheight]{syntaxtree2}
\end{textblock*}
}

\end{frame}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Information Extraction: RE: Knowledge-engineering}

\textbf{Strengths}
\begin{itemize}
\item High precision
\item Interpretable results \ra\ system behaviour is human-comprehensible
\end{itemize}

\textbf{Weaknesses}
\begin{itemize}
\item The writing of rules has no end
\item New rules needed for every new domain 
\begin{itemize}
	\item pattern action rules for shallow approaches
	\item transduction rules for deep approaches
\end{itemize}
\end{itemize}

\end{frame}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Information Extraction: Relation Extraction: Overview}

\gray{Relation Extraction}
\begin{itemize}
\item \gray{Task definition}
\end{itemize}

\textbf{Approaches to Relation Extraction}
\begin{itemize}
\item \gray{Knowledge-engineering approaches to RE}
\item \textbf{Supervised learning approaches to RE}
\item Bootstrapping Approaches to RE
\item Distant Supervision Approaches to RE
\end{itemize}

\end{frame}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Information Extraction: RE: Supervised learning}

\textbf{Question: what is to be learned?}

Answer 1: \myemph{rules}
\begin{itemize}
\item to match \textbf{relation bearing sentences}
\item capture the \textbf{relation arguments} in the matched text
\end{itemize}

Answer 2: \myemph{binary classifier}
\begin{itemize}
\item Classifies a sentence as to whether it bears a specific relation between some entity types
\item Specialized binary classifier
\item Can be divided in two stages: 
\begin{itemize}
\item \myemph{relation detection}: determines whether a sentence expresses a relation (binary)
\item \myemph{relation classification}: determines the relation (multi-way)
\end{itemize}

\item Rule learning popular until early 2000's. Then classifier approach.
\item[\ra] Details on classifier approach only
\end{itemize}

\end{frame}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Information Extraction: RE: Supervised Learning}

Classification approaches to relation extraction:
\begin{itemize}
\item Assume entities  are already tagged \ra\ output of NER
\item Use an algorithm to learn binary classifiers to distinguish instances where
\begin{itemize}
\item entities co-occur and relation holds (positive instances)
\item entities co-occur and relation does not hold (negative instances)
\end{itemize}
\end{itemize}

Key issue: what \textbf{features} to use to represent instances?
\begin{itemize}
\item 3 broad classes:
\begin{itemize}
\item Features of the Named Entities
\item Features from the words in the text
\begin{itemize}
	\item words between the two NE 
	\item words surrounding the candidates (left of 1st word and right of 2nd word)
\end{itemize}
\item Features about the entity pair within the sentence, e.g.
\begin{itemize}
	\item distance between the entities (in words or constituents)
	\item is there a NE in between them?
	\item clues from the syntactic structure of the sentence (parse tree)
\end{itemize}
\end{itemize}

\end{itemize}

\end{frame}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Information Extraction: RE: Supervised Learning: Example}

Ex.: \annot{American Airlines}{ORG}, a unit of \annot{AMR Inc.}{ORG}, immediately matched the move, spokesman \annot{Tim Wagner}{PER} said. \source{Jurafsky and Martin, 2nd ed., p. 730}

Features extracted from this example:
\scriptsize{
\begin{center}
\setlength\extrarowheight{-3pt}
\begin{tabular}{ll}
\toprule
\textbf{Features} & \textbf{Value}  \\ 
\midrule
\MC{2}{l}{\textbf{Entity-based features} }\\
~~~~~~~~Entity$_1$ type			& ORG\\
~~~~~~~~Entity$_1$ head 	& \textit{airlines}\\
~~~~~~~~Entity$_2$ type		& PERS \\
~~~~~~~~Entity$_2$ head 		& \textit{Wagner}\\
~~~~~~~~Concatenated types		& ORGPERS \\ %\midrule
\MC{2}{l}{\textbf{Word-based features} }\\
~~~~~~~~Between-entity BOW 	& \{ \textit{a, unit, of, AMR, Inc., immediately, matched, the, move, spokesman} \}\\
~~~~~~~~Word(s) before Entity$_1$ 	& NONE\\
~~~~~~~~Word(s) before Entity$_2$		& \textit{said}\\ %\midrule
\MC{2}{l}{\textbf{Syntactic features} }\\
~~~~~~~~Constituent path	& NP \ua\ NP \ua\ S \ua\ S \da\ NP\\
~~~~~~~~Base syntactic chunk path & NP \ra\ NP \ra\ PP \ra\ NP \ra\ VP \ra\ NP \ra\ NP \\
~~~~~~~~Typed-dependency path & \textit{Airlines} \la\ $_{subj}$ \textit{matched} \la\ $_{comp}$ \textit{said} \ra\ $_{subj}$ \textit{Wagner} \\
\bottomrule
\end{tabular}

\source{Jurafsky and Martin, 2nd ed., p. 730}
\end{center}
}

\end{frame}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Information Extraction: RE: Supervised Learning}

\textbf{Strengths}
\begin{itemize}
\item No need to write extensive/complex rule sets for each domain
\item System can adapt to any new domain \ra\ provided that training data is supplied
\item[\ra] beware of data sparsity!
\end{itemize}

\textbf{Weaknesses}
\begin{itemize}
\item Quality of relation extraction dependent on quality and quantity of training data
\item[\ra] can be difficult, costly and time consuming to generate
\item Feature extractors can be noisy (e.g. parsers) \ra\ reduce overall performance
\end{itemize}

\end{frame}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Information Extraction: Relation Extraction: Overview}

\gray{Relation Extraction}
\begin{itemize}
\item \gray{Task definition}
\end{itemize}

\textbf{Approaches to Relation Extraction}
\begin{itemize}
\item \gray{Knowledge-engineering approaches to RE}
\item \gray{Supervised learning approaches to RE}
\item \textbf{Bootstrapping Approaches to RE}
\item Distant Supervision Approaches to RE
\end{itemize}

\end{frame}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Information Extraction: Relation Extraction: Bootstrapping}

\textbf{Motivation}: reduce the number of manually labelled examples needed to build a system

\textbf{Requirements}: 
\begin{enumerate}
\item a document collection $\mathcal{D}$
\item set of trusted tuples \textbf{T}, also called \myemph{seed tuples}
\item set of trusted patterns \textbf{P}, also called \myemph{seed patterns}
\end{enumerate}

\textbf{Principle}:
\begin{enumerate}
%\setcounter{enumi}{3}
\item Find tuples from \textbf{T} in $\mathcal{D}$ \Ra\ extract patterns, add them to \textbf{P}
\item Match patterns \textbf{P} in $\mathcal{D}$ \Ra\ extract tuples, add them to \textbf{T}
\item[\ra] Rinse, repeat
\end{enumerate}

\end{frame}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Information Extraction: Relation Extraction: Bootstrapping}

\begin{center}
\includegraphics[width=.55\textwidth]{pattern_based_RE}

\source{Jurafsky and Martin, 2nd ed., p. 740}
\end{center}

\end{frame}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Information Extraction: Relation Extraction: Bootstrapping - DIPRE}

\myemphb{DIPRE} Dual Iterative Pattern Relation Expansion – proposed by Sergie Brin (1999)

\vfill

\textbf{Aim}: to extract useful relational tuples from the Web, of the form (\textsc{person}, \textsc{book\_title})
Ex.: (\textsc{Leo Tolstoy}, \textsc{War and Peace})

\vfill

\textbf{Method}:
\begin{itemize}
\item Exploit duality of patterns and relations
\begin{itemize}
\item Good tuples help find good patterns
\item Good patterns help find good tuples
\end{itemize}

\item Use the bootstrapping method starting with user-supplied tuples
\end{itemize}

\end{frame}

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Information Extraction: Relation Extraction: Bootstrapping - DIPRE}

Main loop in DIPRE is the following:
\begin{enumerate}
\item $\mathcal{R}'$ \la\ Sample
\begin{itemize}
\item[] $\mathcal{R}'$ is a set of tuples \ra\ approximation of target relation
\item[] Sample is the seed tuples (e.g. 5 author-title pairs)
\end{itemize}

\item $\mathcal{O}$ \la\ FindOccurrences($\mathcal{R}'$, $\mathcal{D}$)
\begin{itemize}
\item[] $\mathcal{O}$ contains all occurrences of tuples in $\mathcal{R}'$ appearing in $\mathcal{D}$
\end{itemize}

\item $\mathcal{P}$ \la\ GenPatterns($\mathcal{O}$)
\begin{itemize}
\item[] $\mathcal{P}$ contains the patterns generated based on the occurrences
\item[] Seek for low error rate patterns, ideally having high coverage
\end{itemize}

\item $\mathcal{R}'$ \la\ $M_{\mathcal{D}}(\mathcal{P})$
\begin{itemize}
\item[] Update $\mathcal{R}'$ with the tuples from $\mathcal{D}$ matching patterns in $\mathcal{P}$
\end{itemize}

\item Stop if $\mathcal{R}'$ is large enough, otherwise go to \circled{2}{black}

\end{enumerate}

\end{frame}