Document 23222225

Document technical information

Format pdf
Size 6.5 MB
First found May 22, 2018

Document content analysis

Category Also themed
not defined
no text concepts found





Advanced Language Technologies
Professor Claire Cardie & Professor Lillian Lee
“I’m sorry, Dave, I’m afraid I can’t do that”:
Can computers really understand what we say?
the dream of language technologies
Why is this man smiling?
I’m sorry, Dave, I’m afraid I can’t do that.
Turing predicted we’d be close in about 50 years.
The Turing test:
Intelligence è human-level language use
Do authors dream of electric speech?
“Jarvis”, the A.I.
system in Iron Man
Why is this man not smiling?
Open the pod bay doors, Hal.
from sci-fi to science and engineering
Natural-language processing (NLP)
Recently deployed (in beta): Siri
Goal: create systems that use human language as input/output
•  speech-based interfaces
•  information retrieval / question answering
•  automatic summarization of news, emails, postings, etc.
•  automatic translation
… and much more!
Interdisciplinary: computer science; linguistics, psychology,
communication; probability & statistics, information theory…
Why is this woman smiling?
Credit: AP Photo/Jeopardy Productions Inc.
State of the art: Watson
The Watson system beat human Jeopardy! champions (and
didn’t have internet access; it learned by “reading”
before the match)
Real-life error (1)
But we’re not all the way there yet
A bunch of grapes.
istock | blankaboskov
Hey bunch of grapes
istock | blankaboskov
We can email you when we're back.
Real-life error (3)
Real-life error (2)
[This U.S. city’s] largest airport … We can email you when you're fat.
What is Toronto???
Challenge: ambiguity
List all flights on Tuesday
why is understanding language so hard?
List all flights on Tuesday = List all the flights leaving on Tuesday.
List all flights on Tuesday = Wait ‘til Tuesday, then list all flights.
More realistic example
Baroque example
Retrieve all the local patient files
I saw her duck with a telescope.
Baroque example
Conversation complications
Q: Do you know when the train to Boston leaves?
I saw her duck with a telescope.
A: Yes.
Q: I want to know when the train to Boston leaves.
A: I understand.
[Grishman 1986]
I’m sorry, Dave, I’m afraid I can’t do that.
Meeting these challenges: a brief history
I’m afraid you might be right.
1940s – 50s: From language to probability
Language, statistics, cryptography
“The fundamental problem of communication is that of
reproducing at one point either exactly or approximately a
message selected at another point ...
WWII: Turing helps
break the German
“Enigma” code
[The] semantic aspects of communication are irrelevant to the
engineering problem. The significant aspect is that the actual message is one selected
from a set of possible messages.”
--C. Shannon, 1948
Why is this man smiling?
(An original Enigma machine for encrypting messages is
on display now in the Kroch Library in Olin.) Two probabilities to infer
I can see Alaska from my house!
I can see Alaska from my house!
Prob. of generating this
original message?
Encryption process
Encryption process
Prob. of doing this
encryption of the
[W. Weaver memo on translation, 1949]
Another use of message probs:
speech recognition
1950s-1980s: Breaking with statistics
N. Chomsky (1957):
(a) Colorless green ideas sleep furiously
(1) It’s hard to recognize speech
(2) It’s hard to wreck a nice beach
Both messages have almost the same acoustics, but
different likelihoods.
1990s: The empiricists strike back
•  Huge amounts of data start coming online
•  Advances in algorithms and computational power
“Every time I fire a linguist, my [system’s] performance
goes up” -- F. Jelinek (apocryphal)
(b) Furiously sleep ideas green colorless
The argument: Neither sentence has ever occurred in
the history of English. So any statistical model would
given them the same probability (zero).
The field moved to sophisticated non-probabilistic models
of language.
2000s and beyond: integrating language insights and
statistical techniques
Is Snooki on stork watch?
(wondered in March 2012)
[All 8 results were from March 2011 or earlier] 8
Integrating lang and stats (cont)
Is Snooki on stork watch?
Snooki and fiancé Jionni LaValle are expecting their first child together
the game-changers: Bowie & Iman On Stork Watch
•  data-driven approaches
Monday, February 14, 2000
Rock legend David Bowie and supermodel Iman said yesterday they're expecting their first child
•  models of language
Angie Harmon on Stork Watch
By Marcus Errico
Angie Harmon's going from assistant district attorneying to diaper duty.
The former Law & Order legal dish is expecting her first child with football stud hubby
Jason Sehorn, her publicist confirmed Tuesday.
Why is this man smiling?
What topics might we cover?
Information retrieval
We may hope that machines will eventually compete with men in all purely intellectual fields. But which are the best ones to start with? Even this is a difficult decision.... I do not know what the right answer is, but I think [different] approaches should be tried. Part-of-speech tagging
Text categorization
Word sense disambiguation
Language models
Topic models
Question-answering systems
Semantic analysis Discourse processing
We can only see a short distance ahead, but we can see plenty there that needs to be done. NL generation
Machine translation
Coreference analysis
Dialog systems
Prereqs, Coursework and
Optional textbook: An AI course or permission of instructor
Reference Material
Jurafsky and Martin, Speech and Language Processing, PrenticeHall, 2nd edition. 40%: semester project
problem description and summary of related work (5%), short presentation in class on the planned project
(5%), progress report 1 (2.5%), progress report 2 (2.5%), in-class presentation (10%), final report (15%)
29%: 1 or more research paper presentations, graduate-researcher quality
20%: one-page critiques of research papers, 1 or 2 per class
Other useful references:
•  Manning and Schutze. Foundations of Statistical NLP, MIT
10%: participation
You'll be expected to participate in class discussion or otherwise demonstrate an interest in the material
studied in the course.
1%: course evaluation completion
Press, 1999. •  Others listed on course web page…
Some related courses
This fall:
Computational linguistics (CS3740/LING4424)
Information retrieval (CS/IS4300)
Machine learning (CS4780/5780)
Computational psycholinguistics (PSYCH/LING6280)
Next spring:
(CS4740/5740) Intro NLP
Next year?
NLP and social interaction (CS6742)
Language and technology (INFO4500/6500)

Report this document