LING 438538 Computational Linguistics

1
LING 438/538Computational Linguistics

• Sandiway Fong
• Lecture 21 11/7

2
Administrivia

• Short Lecture Today
• Homework 5
• out today
• due next Tuesday
• usual rules

3
Homework 5
4
Britney Spears

• Webpage
• http//news.bbc.co.uk/cbbcnews/hi/music/newsid_195
  3000/1953614.stm

5

• from the BBC website
• typos 2 is wrong,
• 3 and 5 shouldnt be the same

use this list for the homework
http//www.google.com/jobs/britney.html

• Brittany
• Brittney
• Britany
• Britny
• Briteny
• Britteny
• Briney
• Brittny
• 9. Brintey
• 9. Britanny

6
Question1 Britney Spears

• Question 1
• Part 1 (3pts)
• Compute the edit distances for the misspellings
  of Britney (Spears)
  
• use insertdelete1, substitute2
• Part 2 (3pts)
• Compute the edit distances for the misspellings
  of Britney (Spears)
  
• use insertdelete1, substitute1
• Part 3 (4pts)
• Come up with a metric that correctly ranks the
  top 7 misspellings for either Part 1 or Part 2

7
Making Money from Misspellings

• Webpage
• http//news.bbc.co.uk/1/hi/sci/tech/1575060.stm

8
Making Money from Misspellings

• Excerpts
• US legal authorities are appealing for help in
  tracking down John Zuccarini, who they say is
  making more than a million dollars a year from a
  collection of misspelled domain names.
• The Federal Trade Commission is now looking for
  ways to recover the cash Mr Zuccarini has made
  from the domain names.

• Excerpts
• Mr Zuccarini has been practising a novel
  variation of cybersquatting which usually
  involves gaining control of a website that you
  have no real claim to, and then offering it for
  sale to the rightful owner at a premium.
• The domains registered by Mr Zuccarini were
  typically misspellings of well-known names. Mr
  Zuccarini has reportedly registered 15 variations
  of the spelling of Cartoon Network TV channel,
  and 41 of pop star Britney Spears.

9
Question 2
10
Corpus

• homework corpus
• WSJ9_041.txt
• from the course homepage
• Wall Street Journal articles (July 2628 1989)
• this is the text file you will use
• contains almost 22,000 lines and 150,000 words
• use only the text between the SGML markers
• example...

11
Question 2

• Sun Microsystems Inc. said it will post a
  larger-than-expected fourth-quarter loss of as
  much as 26 million and may show a loss in the
  current first quarter, raising further troubling
  questions about the once high-flying computer
  workstation maker.
• Sun reported last month that management
  errors, rather than a weakness in the market for
  computer workstations, would result in lower
  earnings or a "slight loss" in the quarter ended
  June 30.
• But the amount cited in yesterday's disclosure
  was far greater than analysts had suspected, and
  suggested deepening troubles.
  
• "It is extremely disconcerting," said Peter
  Rogers, computer analyst at Robertson Stephens
  amp Co. in San Francisco.
  
• "Many of us had been led to believe that most (of
  the management-systems problems) had been put
  behind them.
  
• It looks like there is another layer."
• "On the surface, it would lead one to conclude
  that Sun has at least temporarily completely lost
  control of its operations," Mr. Rogers added.
  
• The maker of high-performance desktop
  computers now says the loss was probably between
  20 million and 26 million, compared with
  year-ago net income of 25.3 million, or 66 cents
  a share.
• The huge fourth-quarter loss will bring
  year-end earnings to between 55 million and 61
  million, or between 72 cents and 78 cents a
  share, compared with year-ago net of 66.4
  million, or 89 cents a share.
• Sun said it expects to report fourth-quarter
  revenue of 425 million to 435 million, up 16
  to 19 from a year earlier.
  
• That would contrast sharply with Sun's third
  quarter, when revenue surged 92, and would put
  full-year revenue in the 1.75 billion to 1.77
  billion range, up from 1.05 billion a year
  earlier.

12
Question 2

• Sun said the problems that led to the loss have
  "largely been resolved," and that it received
  record bookings in its fourth quarter.
  
• Still, Sun said profitability in the current
  quarter, ending Sept. 30, can't be assured.
  
• The company added that a return to profitability
  will depend on the effectiveness of cost-cutting
  measures and its ability to obtain parts.
  
• A spokeswoman said the company still faces a
  shortage of certain parts.
  
• In June, Sun said operations were disrupted by
  a change to a new system for getting information
  to management.
  
• The company also cited faulty forecasting of
  demand, problems in manufacturing new machines
  and a shortage of certain parts.
  
• The spokeswoman reiterated that the company sees
  strong demand for its products, and believes the
  market for computer workstations remains healthy.
  
• Sun said it has imposed a hiring freeze in all
  areas except sales and customer service,
  postponed moving into new facilities and
  curtailed other expenses.
• The spokeswoman wouldn't say how much the
  cost-cutting measures will save.
  
• The announcement was made after the market
  closed.
  
• Sun's stock closed at 16.25, up 62.5 cents, in
  national over-the-counter trading.
  

13
Question 2

• edit out other stuff...
• WSJ890728-0079
• 890728
• 890728-0079.
• Major Deficit
• _at_ Signaled by Sun
• _at_ Microsystems
• _at_ ---
• _at_ Firm to Post Quarterly Loss
• _at_ As Much as 26 Million
• _at_ Deepening Trouble Seen
• _at_ ----
• _at_ By Carrie Dolan
• _at_ Staff Reporter of The Wall Street Journal
  
  
• 07/28/89
• WALL STREET JOURNAL (J)
• SUNW
• COMPUTERS AND INFORMATION TECHNOLOGY (CPR)
  
  

14
(Ngram Statistics Package) NSP

• for homework question 2
• suggest you use the NSP software package,
• brew your own, or
• any other package you want to use...

• (Ngram Statistics Package) NSP
• Ted Petersens Perl-based Ngram Statistics
  Package (NSP)
  
• http//www.d.umn.edu/tpederse/nsp.html
• you need to install a free Perl on your system if
  not already available
  
• e.g. Active State Perl

15
(Ngram Statistics Package) NSP

• you only need to use the Perl program file
• count.pl
• NSP on Windows
• command line options

• perl count.pl --help
• Usage count.pl OPTIONS DESTINATION SOURCE ,
  SOURCE ...
  
• Counts up the frequency of all n-grams occurring
  in SOURCE.
  
• Sends to DESTINATION the list of n-grams found,
  along with the
  
• frequencies of combinations of the n tokens that
  the n-gram is composed of. If SOURCE is a
  directory, all text files in it are counted.
  
• OPTIONS
• --ngram N Creates n-grams of N tokens
  each. N 2 by default.
  
• --newLine Prevents n-grams from spanning
  across the new-line character.

16
Question 2

• (4pts)
• List the most frequent closed-class word (for
  each class) in the corpus
  
• use the definition of closed-classes (and your
  judgment) listed in section 8.1 of the textbook
  
• (2pts)
• What is the most frequent proper noun?
• What is the most frequent (non-auxiliary) verb?

17
Question 2

• (8pts)
• compute the probability of the (similar)
  sentences
  
• Bristol-Myers agreed to merge with Sun
  Microsystems
  
• Bristol-Myers and Sun Microsystems agreed to
  merge
  
• using both the bigram and trigram approximations
• use add-one smoothing where relevant

18
Question 2

• Note
• given the chain rule
• p(w1 w2 w3...wn) p(w1) p(w2w1) p(w3w1w2)...
  p(wnw1...wn-2 wn-1)
  
• what is w1 ?
• if were talking about a sentence, w1 START
• Example
• sentence begin with Sun ..
• (see opposite column)
• p(SunSTART)
• assume
• p(START) 1

• Note
• Petersens program does not take into account
  START
  
• youll have to calculate this separately or
  modify the corpus before running NSP...
  
• sentence start symbol START
• file
• START Sun Microsystems Inc. said it will post a
  larger-than-expected fourth-quarter loss of as
  much as 26 million and may show a loss in the
  current first quarter, raising further troubling
  questions about the once high-flying computer
  workstation maker.

19
Summary

• for both 438/538
• Question 1 10pts
• Question 2 14pts
• Total 24 pts