CS447: Natural Language Processing
http://courses.engr.illinois.edu/cs447
Julia Hockenmaier
juliahmr@illinois.edu
3324 Siebel Center
Lecture 3:
Language models
CS447: Natural Language Processing (J. Hockenmaier)
Last lecture’s key concepts
Morphology (word structure): stems, affixes
Derivational vs. inflectional morphology
Compounding
Stem changes
Morphological analysis and generation
Finite-state automata
Finite-state transducers
Composing finite-state transducers
�2
CS447: Natural Language Processing (J. Hockenmaier)
Finite-state transducers
-FSTs define a relation between two regular
languages.
-Each state transition maps (transduces) a character
from the input language to a character (or a
sequence of characters) in the output language
-By using the empty character (ε), characters can
be deleted (x:ε) or inserted(ε:y)
-FSTs can be composed (cascaded), allowing us to
define intermediate representations.
�3
x:y
x:ε ε:y
CS447: Natural Language Processing (J. Hockenmaier)
Today’s lecture
How can we distinguish word salad, spelling errors
and grammatical sentences?
Language models define probability distributions
over the strings in a language.
N-gram models are the simplest and most common
kind of language model.
We’ll look at how they’re defined, how to estimate
(learn) them, and what their shortcomings are.
We’ll also review some very basic probability theory.
�4
CS447: Natural Language Processing (J. Hockenmaier)
Why do we need language models?
Many NLP tasks return output in natural language:
-Machine translation
-Speech recognition
-Natural language generation
-Spell-checking
Language models define probability distributions
over (natural language) strings or sentences.
We can use them to score/rank possible sentences:
If PLM(A) > PLM(B), choose sentence A over B
�5 CS447: Natural Language Processing (J. Hockenmaier)
Reminder:
Basic Probability
Theory
�6
CS447: Natural Language Processing (J. Hockenmaier) �7
P( ) = 2/15
P(blue) = 5/15
P(blue | ) = 2/5
P( ) = 1/15
P(red) = 5/15
P( ) = 5/15
P( or ) = 2/15
P( |red) = 3/5
Pick a random shape, then put it back in the bag.
Sampling with replacement
CS447: Natural Language Processing (J. Hockenmaier) �8
Pick a random shape, then put it back in the bag.
What sequence of shapes will you draw?
P( )
P( )
= 1/15 × 1/15 × 1/15 × 2/15
= 2/50625
= 3/15 × 2/15 × 2/15 × 3/15
= 36/50625
P( ) = 2/15
P(blue) = 5/15
P(blue | ) = 2/5
P( ) = 1/15
P(red) = 5/15
P( ) = 5/15
P( or ) = 2/15
P( |red) = 3/5
Sampling with replacement
CS447: Natural Language Processing (J. Hockenmaier)
Alice was beginning to get very tired of
sitting by her sister on the bank, and of
having nothing to do: once or twice she
had peeped into the book her sister was
reading, but it had no pictures or
conversations in it, ‘and what is the use
of a book,’ thought Alice ‘without
pictures or conversation?’
Alice was beginning to get very tired of
sitting by her sister on the bank, and of
having nothing to do: once or twice she
had peeped into the book her sister was
reading, but it had no pictures or
conversations in it, ‘and what is the use
of a book,’ thought Alice ‘without
pictures or conversation?’
�9
P(of) = 3/66
P(Alice) = 2/66
P(was) = 2/66
P(to) = 2/66
P(her) = 2/66
P(sister) = 2/66
P(,) = 4/66
P(‘) = 4/66
Sampling with replacement
CS447: Natural Language Processing (J. Hockenmaier)
P(of) = 3/66
P(Alice) = 2/66
P(was) = 2/66
P(to) = 2/66
P(her) = 2/66
P(sister) = 2/66
P(,) = 4/66
P(‘) = 4/66
beginning by, very Alice but was and?
reading no tired of to into sitting
sister the, bank, and thought of without
her nothing: having conversations Alice
once do or on she it get the book her had
peeped was conversation it pictures or
sister in, ‘what is the use had twice of
a book”pictures or’ to
�10
In this model, P(English sentence) = P(word salad)
Sampling with replacement
CS447: Natural Language Processing (J. Hockenmaier)
Probability theory: terminology
Trial:
Picking a shape, predicting a word
Sample space Ω:
The set of all possible outcomes
(all shapes; all words in Alice in Wonderland)
Event ω ⊆ Ω:
An actual outcome (a subset of Ω)
(predicting ‘the’, picking a triangle)
�11 CS447: Natural Language Processing (J. Hockenmaier)
Kolmogorov axioms:
1) Each event has a probability between 0 and 1.
2) The null event has probability 0.
The probability that any event happens is 1.
3) The probability of all disjoint events sums to 1.
The probability of events
�12
0 ⇥ P (� � �) ⇥ 1
P (⇤) = 0 and P (�) = 1
�
�i��
P (�i) = 1 if ⇥j �= i : �i ⌅ �j = ⇤
and
�
i �i = �
CS447: Natural Language Processing (J. Hockenmaier)
Bernoulli distribution (two possible outcomes)
The probability of success (=head,yes)
The probability of head is p.
The probability of tail is 1−p.
Categorical distribution (N possible outcomes)
The probability of category/outcome ci is pi
(0≤ pi ≤1 ∑i pi = 1)
Discrete probability distributions:
single trials
�13 CS447: Natural Language Processing (J. Hockenmaier)
The conditional probability of X given Y, P(X | Y),
is defined in terms of the probability of Y, P( Y ),
and the joint probability of X and Y, P(X,Y):
Joint and Conditional Probability
P (X|Y ) =
P (X, Y )
P (Y )
P(blue | ) = 2/5
�14
CS447: Natural Language Processing (J. Hockenmaier) �15
Alice was beginning to get very tired of
sitting by her sister on the bank, and of
having nothing to do: once or twice she
had peeped into the book her sister was
reading, but it had no pictures or
conversations in it, ‘and what is the use
of a book,’ thought Alice ‘without
pictures or conversation?’
P(wi+1 = of | wi = tired) = 1
P(wi+1 = of | wi = use) = 1
P(wi+1 = sister | wi = her) = 1
P(wi+1 = beginning | wi = was) = 1/2
P(wi+1 = reading | wi = was) = 1/2
P(wi+1 = bank | wi = the) = 1/3
P(wi+1 = book | wi = the) = 1/3
P(wi+1 = use | wi = the) = 1/3
Conditioning on the previous word
CS447: Natural Language Processing (J. Hockenmaier) �16
English
Alice was beginning to get very
tired of sitting by her sister on
the bank, and of having nothing to
do: once or twice she had peeped
into the book her sister was
reading, but it had no pictures or
conversations in it, ‘and what is
the use of a book,’ thought Alice
‘without pictures or conversation?’
Word Salad
beginning by, very Alice but was and?
reading no tired of to into sitting
sister the, bank, and thought of without
her nothing: having conversations Alice
once do or on she it get the book her had
peeped was conversation it pictures or
sister in, ‘what is the use had twice of
a book”pictures or’ to
Now, P(English) ⪢ P(word salad)
P(wi+1 = of | wi = tired) = 1
P(wi+1 = of | wi = use) = 1
P(wi+1 = sister | wi = her) = 1
P(wi+1 = beginning | wi = was) = 1/2
P(wi+1 = reading | wi = was) = 1/2
P(wi+1 = bank | wi = the) = 1/3
P(wi+1 = book | wi = the) = 1/3
P(wi+1 = use | wi = the) = 1/3
Conditioning on the previous word
CS447: Natural Language Processing (J. Hockenmaier)
The chain rule
The joint probability P(X,Y) can also be expressed in
terms of the conditional probability P(X | Y)
This leads to the so-called chain rule:
�17
P (X, Y ) = P (X|Y )P (Y )
P (X1, X2, . . . , Xn) = P (X1)P (X2|X1)P (X3|X2, X1)….P (Xn|X1, …Xn�1)
= P (X1)
n�
i=2
P (Xi|X1 . . . Xi�1)
CS447: Natural Language Processing (J. Hockenmaier)
Two random variables X and Y are independent if
If X and Y are independent, then P(X | Y) = P(X):
Independence
P (X, Y ) = P (X)P (Y )
P (X|Y ) =
P (X, Y )
P (Y )
=
P (X)P (Y )
P (Y )
(X ,Y independent)
= P (X)
�18
CS447: Natural Language Processing (J. Hockenmaier)
Probability models
Building a probability model consists of two steps:
1. Defining the model
2. Estimating the model’s parameters
(= training/learning )
Models (almost) always make
independence assumptions.
That is, even though X and Y are not actually independent,
our model may treat them as independent.
This reduces the number of model parameters that
we need to estimate (e.g. from n2 to 2n)
�19 CS447: Natural Language Processing (J. Hockenmaier)
Language modeling
with n-grams
�20
CS447: Natural Language Processing (J. Hockenmaier)
A language model over a vocabulary V assigns
probabilities to strings drawn from V*.
Recall the chain rule:
An n-gram language model assumes each word depends
only on the last n−1 words:
Language modeling with N-grams
P (w1…wi) = P (w1)P (w2|w1)P (w3|w1w2)…P (wi|w1…wi�1)
Pngram(w1…wi) := P (w1)P (w2|w1)…P ( wi⇤⇥�⌅
nth word
| wi�n�1…wi�1⇤ ⇥� ⌅
prev. n�1 words
)
�21 CS447: Natural Language Processing (J. Hockenmaier)
N-gram models
Unigram model P (w1)P (w2)…P (wi)
Bigram model P (w1)P (w2|w1)…P (wi|wi�1)
Trigram model P (w1)P (w2|w1)…P (wi|wi�2wi�1)
N-gram model P (w1)P (w2|w1)…P (wi|wi�n�1…wi�1)
N-gram models assume each word (event) depends
only on the previous n−1 words (events).
Such independence assumptions are called
Markov assumptions (of order n−1).
�22
P(wi|w1…wi�1) :⇡ P(wi|wi�n�1…wi�1)
CS447: Natural Language Processing (J. Hockenmaier)
1. Bracket each sentence by special start and end symbols:
Alice was beginning to get very tired…
(We only assign probabilities to strings …)
2. Count the frequency of each n-gram….
C( Alice) = 1, C(Alice was) = 1,….
3. …. and normalize these frequencies to get the probability:
This is called a relative frequency estimate of P(wn | wn−1)
Estimating N-gram models
�23
P (wn|wn�1) =
C(wn�1wn)
C(wn�1)
CS447: Natural Language Processing (J. Hockenmaier)
Start and End symbols … <\s>
Why do we need a start-of-sentence symbol?
This is just a mathematical convenience, since it allows us to
write e.g. P(w1 | ) for the probability of the first word in
analogy to P(wi+1 | wi ) for any other word.
Why do we need an end-of-sentence symbol?
This is necessary if we want to compare the probability of
strings of different lengths (and actually define a probability
distribution over V*).
We include <\s> in the vocabulary V, require that each string
ends in <\s> and that <\s> can only appear at the end of
sentences, and estimate P(wi+1 = <\s> | wi ).
�24
CS447: Natural Language Processing (J. Hockenmaier)
Parameter estimation (training)
Parameters: the actual probabilities
P(wi = ‘the’ | wi-1 = ‘on’) = ???
We need (a large amount of) text as training data
to estimate the parameters of a language model.
The most basic estimation technique:
relative frequency estimation (= counts)
P(wi = ‘the’ | wi-1 = ‘on’) = C(‘on the’) / C(‘on’)
Also called Maximum Likelihood Estimation (MLE)
MLE assigns all probability mass to events that occur
in the training corpus.
�25 CS447: Natural Language Processing (J. Hockenmaier)
How do we use language models?
Independently of any application, we can use a
language model as a random sentence generator
(i.e we sample sentences according to their language model
probability)
Systems for applications such as machine translation,
speech recognition, spell-checking, generation, often
produce multiple candidate sentences as output.
-We prefer output sentences SOut that have a higher probability
-We can use a language model P(SOut) to score and rank these
different candidate output sentences, e.g. as follows:
argmaxSOut P(SOut | Input) = argmaxSOut P(Input | SOut)P(SOut)
�26
CS447: Natural Language Processing (J. Hockenmaier)
Using n-gram models
to generate language
�27 CS447: Natural Language Processing (J. Hockenmaier)
Generating from a distribution
�28
How do you generate text from an n-gram model?
That is, how do you sample from a distribution P(X |Y=y)?
-Assume X has N possible outcomes (values): {x1, …, xN}
and P(X=xi | Y=y) = pi
-Divide the interval [0,1] into N smaller intervals according to
the probabilities of the outcomes
-Generate a random number r between 0 and 1.
-Return the x1 whose interval the number is in.
x1 x2 x3 x4 x5
0 p1 p1+p2 p1+p2+p3 p1+p2+p3+p4 1
r
CS447: Natural Language Processing (J. Hockenmaier)
Generating the Wall Street Journal
�29 CS447: Natural Language Processing (J. Hockenmaier)
Generating Shakespeare
�30
CS447: Natural Language Processing (J. Hockenmaier)
Intrinsic vs Extrinsic Evaluation
How do we know whether one language model
is better than another?
There are two ways to evaluate models:
– intrinsic evaluation captures how well the model captures
what it is supposed to capture (e.g. probabilities)
-extrinsic (task-based) evaluation captures how useful the
model is in a particular task.
Both cases require an evaluation metric that allows us
to measure and compare the performance of different
models.
�31 CS447: Natural Language Processing (J. Hockenmaier)
How do we evaluate models?
Define an evaluation metric (scoring function).
We will want to measure how similar the predictions
of the model are to real text.
Train the model on a ‘seen’ training set
Perhaps: tune some parameters based on held-out data
(disjoint from the training data, meant to emulate unseen data)
Test the model on an unseen test set
(usually from the same source (e.g. WSJ) as the training data)
Test data must be disjoint from training and held-out data
Compare models by their scores (more on this next week).
�32
CS447: Natural Language Processing (J. Hockenmaier)
Intrinsic Evaluation
of Language Models:
Perplexity
�33 CS447: Natural Language Processing (J. Hockenmaier)
Perplexity
Perplexity is the inverse of the probability of the test
set (as assigned by the language model), normalized
by the number of word tokens in the test set.
Minimizing perplexity = maximizing probability!
Language model LM1 is better than LM2
if LM1 assigns lower perplexity (= higher probability)
to the test corpus w1…wN
NB: the perplexity of LM1 and LM2 can only be directly
compared if both models use the same vocabulary.
�34
CS447: Natural Language Processing (J. Hockenmaier)
The inverse of the probability of the test set,
normalized by the number of tokens in the test set.
Assume the test corpus has N tokens, w1…wN
If the LM assigns probability P(w1, …, wi−n) to the test
corpus, its perplexity, PP(w1…wN), is defined as:
A LM with lower perplexity is better because it assigns
a higher probability to the unseen test corpus.
Perplexity
�35
PP (w1…wN ) = P (w1…wN )�
1
N
= N
⇥
1
P (w1…wN )
= N
⇧⌅⌅⇤
N�
i=1
1
P (wi|w1…wi�1)
=def
N
⇧⌅⌅⇤
N�
i=1
1
P (wi|wi�n…wi�1)
PP (w1…wN ) = P (w1…wN )�
1
N
= N
⇥
1
P (w1…wN )
= N
⇧⌅⌅⇤
N�
i=1
1
P (wi|w1…wi�1)
=def
N
⇧⌅⌅⇤
N�
i=1
1
P (wi|wi�n…wi�1)
CS447: Natural Language Processing (J. Hockenmaier)
Given a test corpus with N tokens, w1…wN,
and an n-gram model P(wi | wi−1, …, wi−n+1)
we compute its perplexity PP(w1…wN) as follows:
Perplexity PP(w1…wn)
(Chain rule)
(N-gram
model)
�36
PP (w1…wN ) = P (w1…wN )�
1
N
= N
⇥
1
P (w1…wN )
= N
⇧⌅⌅⇤
N�
i=1
1
P (wi|w1…wi�1)
=def
N
⇧⌅⌅⇤
N�
i=1
1
P (wi|wi�n…wi�1)
PP (w1…wN ) = P (w1…wN )�
1
N
= N
⇥
1
P (w1…wN )
= N
⇧⌅⌅⇤
N�
i=1
1
P (wi|w1…wi�1)
=def
N
⇧⌅⌅⇤
N�
i=1
1
P (wi|wi�n…wi�1)
PP (w1…wN ) = P (w1…wN )�
1
N
= N
⇥
1
P (w1…wN )
= N
⇧⌅⌅⇤
N�
i=1
1
P (wi|w1…wi�1)
=def
N
⇧⌅⌅⇤
N�
i=1
1
P (wi|wi�n…wi�1)= def
N
s
N
’
i=1
1
P(wi|wi�1, …,wi�n+1)
CS447: Natural Language Processing (J. Hockenmaier)
Practical issues
Since language model probabilities are very small,
multiplying them together often yields to underflow.
It is often better to use logarithms instead, so replace
with
�37
PP(w1…wN) =def
N
s
N
’
i=1
1
P(wi|wi�1, …,wi�n+1)
PP(w
1
…wN) =def exp
✓
�
1
N
N
Â
i=1
logP(wi|wi�1, …,wi�n+1
◆
CS447: Natural Language Processing (J. Hockenmaier)
Perplexity and LM order
Bigram LMs have lower perplexity than unigram LMs
Trigram LMs have lower perplexity than bigram LMs
…
Example from the textbook
(WSJ corpus, trained on 38M tokens, tested on 1.5 M tokens,
vocabulary: 20K word types)
�38
Unigram Bigram Trigram
Perplexity 962 170 109
CS447: Natural Language Processing (J. Hockenmaier)
Extrinsic (Task-Based)
Evaluation of LMs:
Word Error Rate
�39 CS447: Natural Language Processing (J. Hockenmaier)
Intrinsic vs. Extrinsic Evaluation
Perplexity tells us which LM assigns a higher
probability to unseen text
This doesn’t necessarily tell us which LM is better for
our task (i.e. is better at scoring candidate sentences)
Task-based evaluation:
-Train model A, plug it into your system for performing task T
-Evaluate performance of system A on task T.
-Train model B, plug it in, evaluate system B on same task T.
-Compare scores of system A and system B on task T.
�40
CS447: Natural Language Processing (J. Hockenmaier)
Originally developed for speech recognition.
How much does the predicted sequence of words
differ from the actual sequence of words in the correct
transcript?
Insertions: “eat lunch” → “eat a lunch”
Deletions: “see a movie” → “see movie”
Substitutions: “drink ice tea”→ “drink nice tea”
Word Error Rate (WER)
WER =
Insertions + Deletions + Substitutions
Actual words in transcript
�41 CS447: Natural Language Processing (J. Hockenmaier)
But….
… unseen test data will contain unseen words
�42
CS447: Natural Language Processing (J. Hockenmaier)
Getting back to
Shakespeare…
�43 CS447: Natural Language Processing (J. Hockenmaier)
Generating Shakespeare
�44
CS447: Natural Language Processing (J. Hockenmaier)
Shakespeare as corpus
The Shakespeare corpus consists of N=884,647 word
tokens and a vocabulary of V=29,066 word types
Shakespeare produced 300,000 bigram types
out of V2= 844 million possible bigram types.
99.96% of possible bigrams don’t occur in the corpus.
Our relative frequency estimate assigns non-zero
probability to only 0.04% of the possible bigrams
That percentage is even lower for trigrams, 4-grams, etc.
4-grams look like Shakespeare because they are Shakespeare!
�45 CS447: Natural Language Processing (J. Hockenmaier)
We estimated a model on 440K word tokens, but:
Only 30,000 word types occur in the training data
Any word that does not occur in the training data
has zero probability!
Only 0.04% of all possible bigrams (over 30K word
types) occur in the training data
Any bigram that does not occur in the training data
has zero probability (even if we have seen both words in
the bigram)
MLE doesn’t capture unseen events
�46
CS447: Natural Language Processing (J. Hockenmaier)
Zipf’s law: the long tail
1
10
100
1000
10000
100000
1 10 100 1000 10000 100000
Fr
eq
ue
nc
y
(lo
g)
Number of words (log)
How many words occur N times?
W
or
d
fre
qu
en
cy
(l
og
-s
ca
le
)
In natural language:
-A small number of events (e.g. words) occur with high frequency
-A large number of events occur with very low frequency
�47
A few words
are very frequent
English words, sorted by frequency (log-scale)
w1 = the, w2 = to, …., w5346 = computer, …
Most words
are very rare
How many words occur once, twice, 100 times, 1000 times?
the r-th most
common word wr
has P(wr) ∝ 1/r
CS447: Natural Language Processing (J. Hockenmaier)
So….
… we can’t actually evaluate our MLE models on
unseen test data (or system output)…
… because both are likely to contain words/n-grams
that these models assign zero probability to.
We need language models that assign some
probability mass to unseen words and n-grams.
We will get back to this on Friday.
�48
CS447: Natural Language Processing (J. Hockenmaier)
To recap….
�49 CS447: Natural Language Processing (J. Hockenmaier)
Today’s key concepts
N-gram language models
Independence assumptions
Relative frequency (maximum likelihood) estimation
Evaluating language models: Perplexity, WER
Zipf’s law
Today’s reading:
Jurafsky and Martin, Chapter 4, sections 1-4 (2008 edition)
Chapter 3 (3rd Edition)
Friday’s lecture: Handling unseen events!
�50