-
Notifications
You must be signed in to change notification settings - Fork 66
/
NLP-PLM.tex
690 lines (557 loc) · 23.7 KB
/
NLP-PLM.tex
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
%\documentclass[mathserif]{beamer}
\documentclass[handout]{beamer}
%\usetheme{Goettingen}
%\usetheme{Warsaw}
\usetheme{Singapore}
%\usetheme{Frankfurt}
%\usetheme{Copenhagen}
%\usetheme{Szeged}
%\usetheme{Montpellier}
%\usetheme{CambridgeUS}
%\usecolortheme{}
%\setbeamercovered{transparent}
\usepackage[english, activeacute]{babel}
\usepackage[utf8]{inputenc}
\usepackage{amsmath, amssymb}
\usepackage{dsfont}
\usepackage{graphics}
\usepackage{cases}
\usepackage{graphicx}
\usepackage{pgf}
\usepackage{epsfig}
\usepackage{amssymb}
\usepackage{multirow}
\usepackage{amstext}
\usepackage[ruled,vlined,lined]{algorithm2e}
\usepackage{amsmath}
\usepackage{epic}
\usepackage{epsfig}
\usepackage{fontenc}
\usepackage{framed,color}
\usepackage{palatino, url, multicol}
%\algsetup{indent=2em}
\newcommand{\factorial}{\ensuremath{\mbox{\sc Factorial}}}
\newcommand{\BIGOP}[1]{\mathop{\mathchoice%
{\raise-0.22em\hbox{\huge $#1$}}%
{\raise-0.05em\hbox{\Large $#1$}}{\hbox{\large $#1$}}{#1}}}
\newcommand{\bigtimes}{\BIGOP{\times}}
\vspace{-0.5cm}
\title{Natural Language Processing \\ Probabilistic Language Models}
\vspace{-0.5cm}
\author[Felipe Bravo Márquez]{\footnotesize
%\author{\footnotesize
\textcolor[rgb]{0.00,0.00,1.00}{Felipe Bravo-Marquez}}
\date{\today}
\begin{document}
\begin{frame}
\titlepage
\end{frame}
\begin{frame}{Overview}
\begin{scriptsize}
\begin{itemize}
\item The language modeling problem
\item Trigram models
\item Evaluating language models: perplexity
\item Estimation techniques:
\begin{enumerate}\scriptsize{
\item Linear interpolation
\item Discounting methods}
\end{enumerate}
\item This slides are based on the course material by Michael Collins: \url{http://www.cs.columbia.edu/~mcollins/cs4705-spring2019/slides/lmslides.pdf}
\end{itemize}
\end{scriptsize}
\end{frame}
\begin{frame}{The Language Modeling Problem}
\begin{scriptsize}
\begin{itemize}
\item We have some (finite) vocabulary, say $\mathcal{V}$ = $\{$the, a, man, telescope, Beckham, two, . . .$\}$
\item We have an (infinite) set of strings, $\mathcal{V}^*$.
\item For example:
\begin{itemize}\scriptsize{
\item the STOP
\item a STOP
\item the fan STOP
\item the fan saw Beckham STOP
\item the fan saw saw STOP
\item the fan saw Beckham play for Real Madrid STOP}
\end{itemize}
\item Where STOP is a special symbol indicating the end of a sentence.
\end{itemize}
\end{scriptsize}
\end{frame}
\begin{frame}{The Language Modeling Problem (Continued)}
\begin{scriptsize}
\begin{itemize}
\item We have a training sample of example sentences in English.
\item We need to "learn" a probability distribution $p$.
\item $p$ is a function that satisfies:
\begin{align*}
\sum_{x\in V^*} p(x) &= 1 \\
p(x) &\geq 0 \quad \text{for all } x \in V^*
\end{align*}
\item Examples of probabilities assigned to sentences:
\begin{align*}
p(\text{the STOP}) &= 10^{-12} \\
p(\text{the fan STOP}) &= 10^{-8} \\
p(\text{the fan saw Beckham STOP}) &= 2 \times 10^{-8} \\
p(\text{the fan saw saw STOP}) &= 10^{-15} \\
\ldots \\
p(\text{the fan saw Beckham play for Real Madrid STOP}) &= 2 \times 10^{-9}
\end{align*}
\end{itemize}
\end{scriptsize}
\end{frame}
\begin{frame}{The Language Modeling Problem (Continued)}
\begin{scriptsize}
\begin{itemize}
\item Idea 1: The model assigns a higher probability to fluent sentences (those that make sense and are grammatically correct).
\item Idea 2: Estimating this probability function from text (corpus).
\item The language model helps text generation models distinguish between good and bad sentences.
\end{itemize}
\end{scriptsize}
\begin{figure}[h]
\includegraphics[scale = 0.4]{pics/lmdiagram.png}
\end{figure}
\end{frame}
\begin{frame}{Why would we want to do this?}
\begin{scriptsize}
\begin{itemize}
\item Speech recognition was the original motivation.
\item Consider the sentences: 1) recognize speech and 2) wreck a nice beach.
\item These two sentences sound very similar when pronounced, making it challenging for automatic speech recognition systems to accurately transcribe them.
\item When the speech recognition system analyzes the audio input and tries to transcribe it, it takes into account the language model probabilities to determine the most likely interpretation.
\item The language model would favor $p$(recognize speech) over $p$(wreck a nice beach).
\item This is because the former is a more common sentence and should occur more frequently in the training corpus.
\end{itemize}
\end{scriptsize}
\end{frame}
\begin{frame}{Why on earth would we want to do this?}
\begin{scriptsize}
\begin{itemize}
\item By incorporating language models, speech recognition systems can improve accuracy by selecting the sentence that aligns better with linguistic patterns and context, even when faced with similar-sounding alternatives.
\item Related problems are optical character recognition, handwriting recognition.
\item Actually, Language Models are useful in any NLP tasks involving the generation of language (e.g., machine translation, summarization, chatbots).
\item The estimation techniques developed for this problem will be VERY useful for other problems in NLP.
\end{itemize}
\end{scriptsize}
\end{frame}
\begin{frame}{Language Models are Generative}
\begin{scriptsize}
\begin{itemize}
\item Language models can generate sentences by sequentially sampling from probabilities.
\item This is analogous to drawing balls (words) from an urn where their sizes are proportional to their relative frequencies.
\item Alternatively, one could always draw the most probable word, which is equivalent to predicting the next word.
\end{itemize}
\end{scriptsize}
\begin{figure}[h]
\includegraphics[scale = 0.6]{pics/lmsampling.png}
\end{figure}
\end{frame}
\begin{frame}{A Naive Method}
\begin{scriptsize}
\begin{itemize}
\item A very naive method for estimating the probability of a sentence is to count the occurrences of the sentence in the training data and divide it by the total number of training sentences ($N$) to estimate the probability.
\item We have $N$ training sentences.
\item For any sentence $x_1, x_2, \ldots, x_n$, $c(x_1, x_2, \ldots, x_n)$ is the number of times the sentence is seen in our training data.
\item A naive estimate: \begin{displaymath}
p(x1,x2,…,xn)=\frac{c(x_1,x_2 \dots,x_n)}{N}
\end{displaymath}
\item Problem: As the number of possible sentences grows exponentially with sentence length and vocabulary size, it becomes increasingly unlikely for a specific sentence to appear in the training data.
\item Consequently, many sentences will have a probability of zero according to the naive model, leading to poor generalization.
\end{itemize}
\end{scriptsize}
\end{frame}
\begin{frame}{Markov Processes}
\scriptsize
\begin{itemize}
\item Consider a sequence of random variables $X_1, X_2, \ldots, X_n$.
\item Each random variable can take any value in a finite set $V$.
\item For now, we assume the length $n$ is fixed (e.g., $n = 100$).
\item Our goal: model $P(X_1 = x_1, X_2 = x_2, \ldots, X_n = x_n)$
\end{itemize}
\end{frame}
\begin{frame}{First-Order Markov Processes}
\scriptsize
\begin{align*}
P(X_1 = x_1, X_2 = x_2, \ldots, X_n = x_n) &= P(X_1 = x_1) \prod_{i=2}^{n} P(X_i = x_i|X_1 = x_1, \ldots, X_{i-1} = x_{i-1}) \\
&= P(X_1 = x_1) \prod_{i=2}^{n} P(X_i = x_i|X_{i-1} = x_{i-1})
\end{align*}
The first-order Markov assumption: For any $i \in \{2, \ldots, n\}$ and any $x_1, \ldots, x_i$,
\[
P(X_i = x_i|X_1 = x_1, \ldots, X_{i-1} = x_{i-1}) = P(X_i = x_i|X_{i-1} = x_{i-1})
\]
\end{frame}
\begin{frame}{Second-Order Markov Processes}
\scriptsize
\begin{align*}
P(X_1 = x_1, X_2 = x_2, \ldots, X_n = x_n) = \\ P(X_1 = x_1) \cdot P(X_2 = x_2|X_1 = x_1) \cdot \prod_{i=3}^{n} P(X_i = x_i|X_{i-2} = x_{i-2}, X_{i-1} = x_{i-1}) \\
= \prod_{i=1}^{n} P(X_i = x_i|X_{i-2} = x_{i-2}, X_{i-1} = x_{i-1}) \\
\quad \text{(For convenience, we assume $x_0 = x_{-1} = *$, where $*$ is a special "start" symbol.)}
\end{align*}
\end{frame}
\begin{frame}{Modeling Variable Length Sequences}
\scriptsize
\begin{itemize}
\item We would like the length of the sequence, $n$, to also be a random variable.
\item A simple solution: always define $X_n = \text{STOP}$, where STOP is a special symbol.
\item Then use a Markov process as before:
\[
P(X_1 = x_1, X_2 = x_2, \ldots, X_n = x_n) = \prod_{i=1}^{n} P(X_i = x_i|X_{i-2} = x_{i-2}, X_{i-1} = x_{i-1})
\]
\item (For convenience, we assume $x_0 = x_{-1} = *$, where $*$ is a special "start" symbol.)
\end{itemize}
\end{frame}
\begin{frame}{Trigram Language Models}
\scriptsize
\begin{itemize}
\item A trigram language model consists of:
\begin{enumerate}\scriptsize{
\item A finite set $V$
\item A parameter $q(w|u, v)$ for each trigram $u, v, w$ such that $w \in V \cup \{\text{STOP}\}$, and $u, v \in V \cup \{*\}$}
\end{enumerate}
\item For any sentence $x_1 \ldots x_n$ where $x_i \in V$ for $i = 1 \ldots (n-1)$, and $x_n = \text{STOP}$, the probability of the sentence under the trigram language model is:
\[
p(x_1 \ldots x_n) = \prod_{i=1}^{n} q(x_i|x_{i-2}, x_{i-1})
\]
\item We define $x_0 = x_{-1} = *$ for convenience.
\end{itemize}
\end{frame}
\begin{frame}{An Example}
\scriptsize
For the sentence \texttt{the dog barks STOP}, we would have:
\[
p(\text{the dog barks STOP}) = q(\text{the}|*, *) \times q(\text{dog}|*, \text{the}) \times q(\text{barks}|\text{the, dog}) \times q(\text{STOP}|\text{dog, barks})
\]
\end{frame}
\begin{frame}{The Trigram Estimation Problem}
\scriptsize
Remaining estimation problem:
\[
q(w_i | w_{i-2}, w_{i-1})
\]
For example:
\[
q(\text{laughs} | \text{the, dog})
\]
A natural estimate (the "maximum likelihood estimate"):
\[
q(w_i | w_{i-2}, w_{i-1}) = \frac{{\text{Count}(w_{i-2}, w_{i-1}, w_i)}}{{\text{Count}(w_{i-2}, w_{i-1})}}
\]
For instance,
\[
q(\text{laughs} | \text{the, dog}) = \frac{{\text{Count}(\text{the, dog, laughs})}}{{\text{Count}(\text{the, dog})}}
\]
\end{frame}
\begin{frame}{Sparse Data Problems}
\scriptsize
A natural estimate (the "maximum likelihood estimate"):
\[
q(w_i | w_{i-2}, w_{i-1}) = \frac{{\text{Count}(w_{i-2}, w_{i-1}, w_i)}}{{\text{Count}(w_{i-2}, w_{i-1})}}
\]
\[
q(\text{laughs} | \text{the, dog}) = \frac{{\text{Count}(\text{the, dog, laughs})}}{{\text{Count}(\text{the, dog})}}
\]
\begin{itemize}
\item Say our vocabulary size is $N = |V|$, then there are $N^3$ parameters in the model.
\item For example, $N = 20,000 \Rightarrow 20,000^3 = 8 \times 10^{12}$ parameters.
\end{itemize}
\end{frame}
\begin{frame}{Evaluating a Language Model: Perplexity}
\scriptsize
\begin{itemize}
\item We have some test data, $m$ sentences: $s_1, s_2, s_3, ..., s_m$
\item We could look at the probability under our model $\prod_{i=1}^{m} p(s_i)$. Or more conveniently, the log probability:
\[
\log \left( \prod_{i=1}^{m} p(s_i) \right) = \sum_{i=1}^{m} \log p(s_i)
\]
\item In fact, the usual evaluation measure is perplexity:
\[
\text{Perplexity} = 2^{-l} \quad \text{where} \quad l = \frac{1}{M} \sum_{i=1}^{m} \log p(s_i)
\]
\item $M$ is the total number of words in the test data
\end{itemize}
\end{frame}
\begin{frame}{Some Intuition about Perplexity}
\scriptsize
\begin{itemize}
\item Say we have a vocabulary $V$, and $N = |V| + 1$, and a model that predicts:
\[
q(w|u, v) = \frac{1}{N} \quad \text{for all } w \in V \cup \{\text{STOP}\}, \text{ for all } u, v \in V \cup \{*\}
\]
\item It's easy to calculate the perplexity in this case:
\[
\text{Perplexity} = 2^{-l} \quad \text{where} \quad l = \log \frac{1}{N} \Rightarrow \text{Perplexity} = N
\]
\item Perplexity can be seen as a measure of the effective "branching factor"
\end{itemize}
\end{frame}
\begin{frame}{Some Intuition about Perplexity}
\scriptsize
\begin{itemize}
\item \textbf{Proof:} Let's asume we have $m$ sentences of length $n$ in the corpus, and $M$ the amount of tokens in the corpus, $M=m \cdot n$.
\item Let's consider the log (base 2) probability of a sentence $s = w_1 w_2 \dots w_n$ under the model:
\[
\log p(s) = \log \prod_{i=1}^{n} q(w_i|w_{i-2}, w_{i-1}) = \sum_{i=1}^{n} \log q(w_i|w_{i-2}, w_{i-1})
\]
\item Since each $q(w_i|w_{i-2}, w_{i-1})$ is equal to $\frac{1}{N}$, we have:
\[
\log p(s) = \sum_{i=1}^{n} \log \frac{1}{N} = n \cdot \log \frac{1}{N} = -n \cdot \log N
\]
\[
l = \frac{1}{M} \sum_{i=1}^{m} \log p(s_i) = \frac{1}{M} \sum_{i=1}^{m} -n \cdot \log N = \frac{1}{M} \cdot -m \cdot n \cdot \log N = - \log N
\]
\item Therefore, the perplexity is given by:
\[
\text{Perplexity} = 2^{-l} = 2^{-(- \log N)} = N
\]
\end{itemize}
\end{frame}
\begin{frame}{Typical Values of Perplexity}
\scriptsize
\begin{itemize}
\item Results from Goodman ("A bit of progress in language modeling"), where $|V| = 50,000$ \cite{goodman2001bit}.
\item A trigram model: $p(x_1, \ldots, x_n) = \prod_{i=1}^n q(x_i | x_{i-2}, x_{i-1})$ \\
Perplexity = 74
\item A bigram model: $p(x_1, \ldots, x_n) = \prod_{i=1}^n q(x_i | x_{i-1})$ \\
Perplexity = 137
\item A unigram model: $p(x_1, \ldots, x_n) = \prod_{i=1}^n q(x_i)$ \\
Perplexity = 955
\end{itemize}
\end{frame}
\begin{frame}{Some History}
\scriptsize
\begin{itemize}
\item Shannon conducted experiments on the entropy of English, specifically investigating how well people perform in the perplexity game.
\item Reference: C. Shannon. ``Prediction and entropy of printed English.'' \textit{Bell Systems Technical Journal}, 30:50–64, 1951. \cite{shannon1951prediction}
\end{itemize}
\begin{figure}[h]
\includegraphics[scale = 0.4]{pics/shannon.png}
\end{figure}
\end{frame}
\begin{frame}{Some History}
\scriptsize
\begin{itemize}
\item Chomsky, in his book \textit{Syntactic Structures} (1957), made several important points regarding grammar. \cite{chomsky2009syntactic}
\item According to Chomsky, the notion of "grammatical" cannot be equated with "meaningful" or "significant" in a semantic sense.
\item He illustrated this with two nonsensical sentences:
\begin{itemize}
\scriptsize
\item (1) Colorless green ideas sleep furiously.
\item (2) Furiously sleep ideas green colorless.
\end{itemize}
\item While both sentences lack meaning, Chomsky argued that only the first one is considered grammatical by English speakers.
\end{itemize}
\end{frame}
\begin{frame}{Some History}
\scriptsize
\begin{itemize}
\item Chomsky also emphasized that grammaticality in English cannot be determined solely based on statistical approximations.
\item Even though neither sentence (1) nor (2) has likely occurred in English discourse, a statistical model would consider them equally "remote" from English.
\item However, sentence (1) is grammatical, while sentence (2) is not, highlighting the limitations of statistical approaches in capturing grammaticality.
\end{itemize}
\begin{figure}[h]
\includegraphics[scale = 0.4]{pics/chomsky.png}
\end{figure}
\end{frame}
\begin{frame}{The Bias-Variance Trade-Off}
\scriptsize
\begin{itemize}
\item Trigram maximum-likelihood estimate:
\[
q_{\text{ML}}(w_i | w_{i-2}, w_{i-1}) = \frac{{\text{Count}(w_{i-2}, w_{i-1}, w_i)}}{{\text{Count}(w_{i-2}, w_{i-1})}}
\]
\item Bigram maximum-likelihood estimate:
\[
q_{\text{ML}}(w_i | w_{i-1}) = \frac{{\text{Count}(w_{i-1}, w_i)}}{{\text{Count}(w_{i-1})}}
\]
\item Unigram maximum-likelihood estimate:
\[
q_{\text{ML}}(w_i) = \frac{{\text{Count}(w_i)}}{{\text{Count()}}}
\]
\end{itemize}
\end{frame}
\begin{frame}{Linear Interpolation}
\scriptsize
\begin{itemize}
\item Take our estimate $q(w_i | w_{i-2}, w_{i-1})$ to be
\[
q(w_i | w_{i-2}, w_{i-1}) = \lambda_1 \cdot q_{\text{ML}}(w_i | w_{i-2}, w_{i-1}) + \lambda_2 \cdot q_{\text{ML}}(w_i | w_{i-1}) + \lambda_3 \cdot q_{\text{ML}}(w_i)
\]
where $\lambda_1 + \lambda_2 + \lambda_3 = 1$, and $\lambda_i \geq 0$ for all $i$.
\item Our estimate correctly defines a distribution (define $V' = V \cup \{\text{STOP}\}$):
\[
\begin{aligned}
& \sum_{w \in V'} q(w | u, v) \\
= & \sum_{w \in V'} [\lambda_1 \cdot q_{\text{ML}}(w | u, v) + \lambda_2 \cdot q_{\text{ML}}(w | v) + \lambda_3 \cdot q_{\text{ML}}(w)] \\
= & \lambda_1 \sum_{w} q_{\text{ML}}(w | u, v) + \lambda_2 \sum_{w} q_{\text{ML}}(w | v) + \lambda_3 \sum_{w} q_{\text{ML}}(w) \\
= & \lambda_1 + \lambda_2 + \lambda_3 = 1
\end{aligned}
\]
\item We can also show that $q(w | u, v) \geq 0$ for all $w \in V'$.
\end{itemize}
\end{frame}
\begin{frame}{Estimating $\lambda$ Values}
\scriptsize
\begin{itemize}
\item Hold out part of the training set as \textit{validation} data.
\item Define $c'(w_1, w_2, w_3)$ to be the number of times the trigram $(w_1, w_2, w_3)$ is seen in the validation set.
\item Choose $\lambda_1$, $\lambda_2$, $\lambda_3$ to maximize:
\[
L(\lambda_1, \lambda_2, \lambda_3) = \sum_{w_1,w_2,w_3} c'(w_1, w_2, w_3) \log q(w_3 | w_1, w_2)
\]
such that $\lambda_1 + \lambda_2 + \lambda_3 = 1$, and $\lambda_i \geq 0$ for all $i$, and where
\[
q(w_i | w_{i-2}, w_{i-1}) = \lambda_1 \cdot q_{\text{ML}}(w_i | w_{i-2}, w_{i-1}) + \lambda_2 \cdot q_{\text{ML}}(w_i | w_{i-1}) + \lambda_3 \cdot q_{\text{ML}}(w_i)
\]
\end{itemize}
\end{frame}
\begin{frame}{Discounting Methods}
\begin{scriptsize}
\begin{itemize}
\item Consider the following counts and maximum-likelihood estimates:
\end{itemize}
\begin{table}[h]
\centering
\begin{tabular}{lll}
\textbf{Sentence} & \textbf{Count} & \textbf{$q_{\text{ML}}(w_i | w_{i-1})$} \\
\hline
the & 48 & \\
the, dog & 15 & $15/48$ \\
the, woman & 11 & $11/48$ \\
the, man & 10 & $10/48$ \\
the, park & 5 & $5/48$ \\
the, job & 2 & $2/48$ \\
the, telescope & 1 & $1/48$ \\
the, manual & 1 & $1/48$ \\
the, afternoon & 1 & $1/48$ \\
the, country & 1 & $1/48$ \\
the, street & 1 & $1/48$ \\
\end{tabular}
\end{table}
\begin{itemize}
\item The maximum-likelihood estimates are high, particularly for low count items.
\end{itemize}
\end{scriptsize}
\end{frame}
\begin{frame}{Discounting Methods}
\begin{scriptsize}
\begin{itemize}
\item Define ``discounted'' counts as follows:
\begin{displaymath}
\text{Count}^*(x) =\text{Count}(x)-0.5
\end{displaymath}
\end{itemize}
\begin{table}[h]
\centering
\begin{tabular}{llll}
\textbf{Sentence} & \textbf{Count} & \textbf{Count}*$\mathbf{(x)}$ & $\mathbf{q_{\text{ML}}(w_i | w_{i-1})}$ \\
\hline
the & 48 & & \\
the, dog & 15 & 14.5 & $14.5/48$ \\
the, woman & 11 & 10.5 & $10.5/48$ \\
the, man & 10 & 9.5 & $9.5/48$ \\
the, park & 5 & 4.5 & $4.5/48$ \\
the, job & 2 & 1.5 & $1.5/48$ \\
the, telescope & 1 & 0.5 & $0.5/48$ \\
the, manual & 1 & 0.5 & $0.5/48$ \\
the, afternoon & 1 & 0.5 & $0.5/48$ \\
the, country & 1 & 0.5 & $0.5/48$ \\
the, street & 1 & 0.5 & $0.5/48$ \\
\end{tabular}
\end{table}
\begin{itemize}
\item The new estimates are based on the discounted counts.
\end{itemize}
\end{scriptsize}
\end{frame}
\begin{frame}{Discounting Methods (Continued)}
\begin{itemize}
\item We now have some "missing probability mass":
\[
\alpha(w_{i-1}) = 1 - \sum_{w} \frac{{\text{{Count}}^*(w_{i-1}, w)}}{{\text{{Count}}(w_{i-1})}}
\]
\item For example, in our case:
\[
\alpha(\text{{the}}) = \frac{{10 \times 0.5}}{{48}} = \frac{{5}}{{48}}
\]
\end{itemize}
\end{frame}
\begin{frame}{Katz Back-Off Models (Bigrams)}
\begin{itemize}
\item For a bigram model, define two sets:
\[
A(w_{i-1}) = \{w : \text{Count}(w_{i-1}, w) > 0\}
\]
\[
B(w_{i-1}) = \{w : \text{Count}(w_{i-1}, w) = 0\}
\]
\item A bigram model:
\[
q_{\text{BO}}(w_i | w_{i-1}) =
\begin{cases}
\frac{\text{Count}^*(w_{i-1}, w_i)}{\text{Count}(w_{i-1})} & \text{if } w_i \in A(w_{i-1}) \\
\frac{\alpha(w_{i-1}) q_{\text{ML}}(w_i)}{\sum_{w \in B(w_{i-1})} q_{\text{ML}}(w)} & \text{if } w_i \in B(w_{i-1})
\end{cases}
\]
\item Where:
\[
\alpha(w_{i-1}) = 1 - \sum_{w \in A(w_{i-1})} \frac{\text{Count}^*(w_{i-1}, w)}{\text{Count}(w_{i-1})}
\]
\end{itemize}
\end{frame}
\begin{frame}{Katz Back-Off Models (Trigrams)}
\begin{scriptsize}
\begin{itemize}
\item For a trigram model, first define two sets:
\[
A(w_{i-2}, w_{i-1}) = \{w : \text{Count}(w_{i-2}, w_{i-1}, w) > 0\}
\]
\[
B(w_{i-2}, w_{i-1}) = \{w : \text{Count}(w_{i-2}, w_{i-1}, w) = 0\}
\]
\item A trigram model is defined in terms of the bigram model:
\[
q_{\text{BO}}(w_i | w_{i-2}, w_{i-1}) =
\begin{cases}
\frac{\text{Count}^*(w_{i-2}, w_{i-1}, w_i)}{\text{Count}(w_{i-2}, w_{i-1})} & \text{if } w_i \in A(w_{i-2}, w_{i-1}) \\
\frac{\alpha(w_{i-2}, w_{i-1}) q_{\text{BO}}(w_i|w_{i-1})}{\sum_{w \in B(w_{i-2}, w_{i-1})} q_{\text{BO}}(w|w_{i-1})} & \text{if } w_i \in B(w_{i-2}, w_{i-1})
\end{cases}
\]
\item Where:
\[
\alpha(w_{i-2}, w_{i-1}) = 1 - \sum_{w \in A(w_{i-2}, w_{i-1})} \frac{\text{Count}^*(w_{i-2}, w_{i-1}, w)}{\text{Count}(w_{i-2}, w_{i-1})}
\]
\end{itemize}
\end{scriptsize}
\end{frame}
\begin{frame}{Summary}
\scriptsize
\begin{itemize}
\item Deriving probabilities in probabilistic language models involves three steps:
\begin{enumerate}\scriptsize{
\item Expand $p(w_1, w_2, \ldots, w_n)$ using the Chain rule.
\item Apply Markov Independence Assumptions \\
$p(w_i | w_1, w_2, \ldots, w_{i-2}, w_{i-1}) = p(w_i | w_{i-2}, w_{i-1})$.
\item Smooth the estimates using low order counts.}
\end{enumerate}
\item Other methods for improving language models include:
\begin{itemize}\scriptsize{
\item Introducing latent variables to represent topics, known as topic models. \cite{blei2003latent}
\item Replacing $p(w_i | w_1, w_2, \ldots, w_{i-2}, w_{i-1})$ with a predictive neural network and an ``embedding layer'' to better represent larger contexts and leverage similarities between words in the context. \cite{bengio2000neural}}
\end{itemize}
\item Modern language models utilize deep neural networks in their backbone and have a vast parameter space.
\end{itemize}
\end{frame}
\begin{frame}
\frametitle{Questions?}
%\vspace{1.5cm}
\begin{center}\LARGE Thanks for your Attention!\\ \end{center}
\end{frame}
\begin{frame}[allowframebreaks]\scriptsize
\frametitle{References}
\bibliography{bio}
\bibliographystyle{apalike}
%\bibliographystyle{flexbib}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%
\end{document}