Language and Learning Introduction to Artificial Intelligence COS302 Michael L. Littman Fall 2001.

Language and LearningLanguage and Learning

Introduction toIntroduction toArtificial IntelligenceArtificial Intelligence

COS302COS302

Michael L. LittmanMichael L. Littman

Fall 2001Fall 2001

AdministrationAdministration

Break ok?Break ok?

Search and AISearch and AI

Powerful techniques. Do they solve Powerful techniques. Do they solve the whole AI problem?the whole AI problem?

Let’s do a thought experiment.Let’s do a thought experiment.

Chinese Room ArgumentChinese Room Argument

Searle: There is a fundamental Searle: There is a fundamental difference between symbol difference between symbol manipulation and understanding manipulation and understanding meaning.meaning.

Syntax vs. semanticsSyntax vs. semantics

Was Searle Right?Was Searle Right?

Yes/no: The richness of corpora.Yes/no: The richness of corpora.

What is judgment? Can it be What is judgment? Can it be automated?automated?

Famous QuotesFamous Quotes

““The difference between chess and The difference between chess and crossword puzzles is that, in chess, crossword puzzles is that, in chess, you know when you’ve won.” --- you know when you’ve won.” --- Michael L. LittmanMichael L. Littman

““Trying is the first stem toward Trying is the first stem toward failure.” --- Homer Simpson via my failure.” --- Homer Simpson via my cryptogram programcryptogram program

Cryptogram ExampleCryptogram Example

Is this wrong?Is this wrong?

Can you write a program that would Can you write a program that would agree with you on this?agree with you on this?

What would your program be like?What would your program be like?

Language ResourcesLanguage Resources

Three major resources:Three major resources:• DictionariesDictionaries• Labeled corporaLabeled corpora• Unlabeled corporaUnlabeled corpora

Each useful for different purposes.Each useful for different purposes.

Examples to follow…Examples to follow…

GoogleGoogle

Word matching on large corpusWord matching on large corpus

Hubs and authorities (unlabeled Hubs and authorities (unlabeled corpus, statistical processing)corpus, statistical processing)

Hand tuned ranking functionHand tuned ranking function

http://www.google.comhttp://www.google.com

Also machine translation…Also machine translation…

IonautIonaut

Question answering: Question answering: www.ionaut.comwww.ionaut.com

Hand-built question categorizationHand-built question categorizationNamed-entity tagger trained from Named-entity tagger trained from

tagged corpustagged corpusLarge unlabeled text corpusLarge unlabeled text corpusHand-tuned ranking rulesHand-tuned ranking rules

Ask JeevesAsk Jeeves

Hand-selected web pages and Hand-selected web pages and corresponding questionscorresponding questions

Proprietary mapping from query to Proprietary mapping from query to question in databasequestion in database

www.ask.comwww.ask.com

NL for DBNL for DB

Hand constructed rules turn Hand constructed rules turn sentences into DB queriessentences into DB queries

STARTSTART

http://www.ai.mit.edu/projects/http://www.ai.mit.edu/projects/infolab/ailab.htmlinfolab/ailab.html

ElizaEliza

Chatterbots very popular. Some Chatterbots very popular. Some believe they can replace “customer believe they can replace “customer care specialists”.care specialists”.

Generally a large collection of rules Generally a large collection of rules and example text.and example text.

http://www.uwec.edu/Academic/http://www.uwec.edu/Academic/Curric/jerzdg/if/WebHelp/eliza.htmCurric/jerzdg/if/WebHelp/eliza.htm

WordnetWordnet

Hand builtHand built

Rich interconnectionsRich interconnections

Showing up as a resource in many Showing up as a resource in many systems.systems.

http://www.cogsci.princeton.edu/cgi-http://www.cogsci.princeton.edu/cgi-bin/bin/

webwnwebwn

Spelling CorrectionSpelling Correction

Semi-automated selection of Semi-automated selection of confusable pairs.confusable pairs.

System trained on large corpus, System trained on large corpus, giving positive and negative giving positive and negative examples (WSJ)examples (WSJ)

http://l2r.cs.uiuc.edu/~cogcomp/eoh/http://l2r.cs.uiuc.edu/~cogcomp/eoh/spelldemo.htmlspelldemo.html

OneAcrossOneAcross

Large corpus of crossword answers: Large corpus of crossword answers: www.oneacross.comwww.oneacross.com

IR-style techniques to find relevant IR-style techniques to find relevant cluesclues

Ranking function trained from held-Ranking function trained from held-out cluesout clues

Learns from usersLearns from users

Essay GradingEssay Grading

Unsupervised learning to discover Unsupervised learning to discover word representationsword representations

Labeled graded essaysLabeled graded essays

http://www.knowledge-http://www.knowledge-technologies.com/IEAdemo.htmltechnologies.com/IEAdemo.html

More ApplicationsMore Applications

Word-sense disambiguationWord-sense disambiguation

Part of speech taggingPart of speech tagging

ParsingParsing

Reading comprehensionReading comprehension

SummarizationSummarization

CobotCobot

Cross-language IRCross-language IR

Text categorizationText categorization

SynonymsSynonyms

Carp = quit, argue, painful, scratch, Carp = quit, argue, painful, scratch, complaincomplain

Latent Semantic IndexingLatent Semantic Indexing• Corpus, deep statistical analysisCorpus, deep statistical analysis

Pointwise Mutual InformationPointwise Mutual Information• HugeHuge corpus, shallow analysis corpus, shallow analysis

WordNet…WordNet…

AnalogiesAnalogies

OvercoatOvercoat::warmthwarmth::::• GloveGlove::handhand• JewelryJewelry::wealthwealth• SlickerSlicker::moisturemoisture• DisguiseDisguise::identificationidentification• HelmetHelmet::protectionprotectionDictionary not sufficientDictionary not sufficientLabeled corpus probably wouldn’t helpLabeled corpus probably wouldn’t helpUnlabeled corpus, not obvious…Unlabeled corpus, not obvious…

What to LearnWhat to Learn

Difference between straight search Difference between straight search problems and languageproblems and language

Why learning might helpWhy learning might help

Three types of resources (hand-Three types of resources (hand-created, labeled, unlabeled)created, labeled, unlabeled)

A RuleA Rule

Follow this carefully.Follow this carefully.

ExplanationExplanation

Homework 5 (due 11/7)Homework 5 (due 11/7)

1.1. The value iteration algorithm from the The value iteration algorithm from the Games of ChanceGames of Chance lecture can be lecture can be applied to deterministic games with applied to deterministic games with loops. Argue that it produces the same loops. Argue that it produces the same answer as the “Loopy” algorithm from answer as the “Loopy” algorithm from the the Game TreeGame Tree lecture. lecture.

2.2. Write the matrix form of the game tree Write the matrix form of the game tree below.below.

Game TreeGame Tree

X-1

+2

-1 +4

Y-2 Y-3

X-4

L R

L R

L R

+5L

+2R

ContinuedContinued

3. How many times (on average) do 3. How many times (on average) do you need to flip a coin before you you need to flip a coin before you flip 3 heads in a row? (a) Set this flip 3 heads in a row? (a) Set this up as a Markov chain, and (b) solve up as a Markov chain, and (b) solve it.it.

Homework 6 (due 11/14)Homework 6 (due 11/14)

1.1. Use the web to find sentences to Use the web to find sentences to support the analogy support the analogy traffic:street::water:riverbed. traffic:street::water:riverbed. Give the sentences and their Give the sentences and their sources.sources.

2.2. More soon…More soon…