ELUTE E ssential L ibraries and U tilities of T ext E ngineering Tian-Jian Jiang.

ELUTEEssential Libraries and Utilities of

Text Engineering

Tian-Jian Jiang

Why bother?Since we already have...

(lib)TaBE•Traditional Chinese Word Segmentation

•with Big5 encoding

•Traditional Chinese Syllable-to-Word Conversion

•with Big5 encoding

•for bo-po-mo-fo transcription system

1999 - 2001?1999 - 2001?1999 - 2001?1999 - 2001?

How about...

libchewing•Now hacking for

•UTF-8 encoding

•Pinyin transcription system

•and looking for

•an alternative algorithm

•a better dictionary

We got a problem•23:43 < s*****> 到底 (3023) + 螫舌 (0) 麼(2536) 東西 (6024) = 11583

•23:43 < s*****> 到底是 (829) + 什麼東西 (337) = 1166

•23:43 < s*****> 到底螫舌麼東西 大勝 到底這什麼東西

•00:02 < s*****> k***: 「什麼」會被「什麼東西」排擠掉

•00:02 < s*****> k***: 結果是 20445 活生生的被 337 幹掉 :P

Word Segmentation Review

Heuristic Rules*

• Maximum matching -- Simple vs. Complex: 下雨天真正討厭

• 下雨 天真 正 討厭 vs. 下雨天 真正 討厭

• Maximum average word length

• 國際化

• Minimum variance of word lengths

• 研究 生命 起源

• Maximum degree of morphemic freedom of single-character word

• 主要 是 因為

* Refer to MMSEG by C. H. Tsai: http://technology.chtsai.org/mmseg/

Graphical Models• Markov chain family

• Statistical Language Model (SLM)

• Hidden Markov Model (HMM)

• Exponential models

• Maximum Entropy (ME)

• Conditional Random Fields (CRF)

• Applications

• Probabilistic Context-Free Grammar (PCFG) Parser

• Head-driven Phrase Structure Grammar (HPSG) Parser

• Link Grammar Parser

What is a language model?

A probability distributionover surface patterns of

texts.

The Italian Who Went to Malta

•One day ima gonna Malta to bigga hotel.

•Ina morning I go down to eat breakfast.

•I tella waitress I wanna two pissis toasts.

•She brings me only one piss.

•I tella her I want two piss. She say go to the toilet.

•I say, you no understand, I wanna piss onna my plate.

•She say you better no piss onna plate, you sonna ma bitch.

•I don’t even know the lady and she call me sonna ma bitch!

P(“I want to piss”) > P(“I want two pieces”)

For that Malta waitress,

Do the Math• Conditional probability:

•

• Bayes’ theorem:

•

• Information theory:

• Noisy channel model

•

• Language model: P(i)

)(

)()|(

BP

BAPBAP

Noisy channelp(o|i)

DecoderI O Î

)(

)()|(

)(

)()|(

AP

BPBAP

AP

ABPABP

)|()(maxarg)(

)|()(maxarg)|(maxarg

^

iopipop

iopipoipI

iii

Shannon’s Game

•Predict next word by history

•

•Maximum Likelihood Estimation

•

•C(w1…wn) : Frequency of n-gram w1…wn

)(

)()|(

11

111

n

nnn wwP

wwPwwwP

)(

)()|(

11

111MLE

n

nnn wwC

wwCwwwP

Once in a Blue Moon

• A cat has seen...

• 10 sparrows

• 4 barn swallows

• 1 Chinese Bulbul

• 1 Pacific Swallow

• How likely is it that next bird is unseen?

(1+1) / (10 + 4 + 1 + 1)

But I’ve seen a moonand I’m blue

• Simple linear interpolation

• PLi(wn|wn-2 , wn-1) = λ1P1(wn) + λ2P2(wn|wn-1 ) + λ3P2(wn|wn-1 , wn-2)

• 0 ≤λi ≤ 1, Σiλi = 1

• Katz’s backing-off

• Back-off through progressively shorter histories.

• Pbo(wi|wi-(n-1)…wi-1) =

•

•kwwC

wwC

wwCd ini

ini

iniww ini

)( if ,)(

)()1( )1(

1)1(

)1(

1)1(

otherwise. ),|( 1)2(bo1)1( iniiww wwwPini

Good Luck!• Place a bet remotely on

a horse race within 8 horses by passing encoded messages.

• Past bet distribution

• horse 1: 1/2

• horse 2: 1/4

• horse 3: 1/8

• horse 4: 1/16

• the rest: 1/64

Foreversoul: http://flickr.com/photos/foreversouls/CC: BY-NC-ND

3 bits? No, only 2!

0, 10, 110, 1110, 111100, 111101, 111110, 111111

Alright, let’s ELUTE

Bi-gram MLE Flow Chart

Permute candidates

right_gramIn LM?

hasleft_gram?

bi_gramIn LM?

left_gramIn LM?

temp_score =LogProb(right_gram)

temp_score =LogProb(bi_gram)

temp_score =LogProb(left_gram) +BackOff(right_gram)

temp_score =LogProb(Unknown) +BackOff(right_gram)

temp_score =LogProb(Unknown)

Update scores

temp_score +=previous_score

have 2 grams?

Yes

Yes

Yes

Yes Yes

No

No

No

No No

INPUT input_syllables; len = Length(input_syllables); Load(language_model);scores[len + 1]; tracks[len + 1]; words[len + 1];FOR i = 0 TO len

scores[i] = 0.0; tracks[i] = -1; words[i] = "";FOR index = 1 TO len

best_score = 0.0; best_prefix = -1; best_word = "";FOR prefix = index - 1 TO 0

right_grams[] = Homophones(Substring(input_syllabes, prefix, index - prefix));FOR EACH right_gram IN right_grams[]

IF right_gram IN language_modelleft = tracks[prefix];IF left >= 0 AND left != prefix

left_grams[] = Homophones(Substring(input_syllables, left, prefix - left));FOR EACH left_gram IN left_grams[]

temp_score = 0.0;bigram = left_gram + " " + right_gram;IF bigram IN language_model

bigram_score = LogProb(bigram);temp_score += bigram_score;

ELSE IF left_gram IN language_modelbigram_backoff = LogProb(left_gram) + BackOff(right_gram);temp_score += bigram_backoff;

ELSEtemp_score += LogProb(Unknown) + BackOff(right_gram);

temp_score += scores[prefix];Scoring

ELSEtemp_score = LogProb(right_gram);Scoring

ELSEtemp_score = LogProb(Unknown) + scores[prefix];Scoring

scores[index] = best_score; tracks[index] = best_prefix_index; words[index] = best_prefix;IF tracks[index] == -1

tracks[index] = index - 1;boundary = len; output_words = "";WHILE boundary > 0

output_words = words[boundary] + output_words;boundary = tracks[boundary];

RETURN output_words;

SUBROUTINE ScoringIF best_score == 0.0 OR temp_score > best_score

best_score = temp_score;best_prefix = prefix;best_word = right_gram;

Bi-gram Syllable-to-Word

Show me the…

William’s Requests

And My Suggestions

• Convenient API

• Plain text I/O (in UTF-8)

• More linguistic information

• Algorithm: CRF

• Corpus: we need YOU!

• Flexible to different applications

• Composite, Iterator, and Adapter Patterns

• IDL support

• SWIG

• Open Source

• Open Corpus, too

Thank YOU