DOC PREVIEW
MIT 6 863J - Lecture 16: the boundaries of syntax & semantics

This preview shows page 1-2-24-25 out of 25 pages.

Save
View full document
View full document
Premium Document
Do you want full access? Go Premium and unlock all 25 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 25 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 25 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 25 pages.
Access to all documents
Download any document
Ad free experience
Premium Document
Do you want full access? Go Premium and unlock all 25 pages.
Access to all documents
Download any document
Ad free experience

Unformatted text preview:

6.863J Natural Language ProcessingLecture 16: the boundaries of syntax & semantics – towards constraint-based systemsRobert C. [email protected]/9.611J Lecture 16 Sp03The Menu Bar• Administrivia:• Lab 4 due April 9? (what about Friday) • Start w/ final projects, unless there are objections• Agenda: • Shallow instead of ‘deep’ semantics: MUC• Stochastic language use? Some examples• How to accommodate: towards constraint-based grammar6.863J/9.611J Lecture 16 Sp03How to integrate all this stuff?• We saw that we might want to partition syntactic knowledge from semantic…• But we have to go farther – because both of these might be probabilistic in nature• How to integrate?6.863J/9.611J Lecture 16 Sp03Integration• One way: semantic grammar (see below)• The way we’ll explore though:• Define linguistic structure• Place pr distributions on that structure6.863J/9.611J Lecture 16 Sp03Example• Syntactic rule = NP à Det N• Semantic extension = NP : Apply(lambda (x) (DEF/SING x), N)• Lexicon:• Art:the: DEF/SING• N:guy: Person• Parse of the NP the guy:• NP à Det N ⇒ OK, NP contains article & noun• Apply (lambda (x) (DEF/SING x), N) ⇒ OK, NP contains DEF/SING article• Apply (lambda (x) (DEF/SING x), Person) ⇒ the N in the NP is Person• (DEF/SING Person) ⇒ result of applying lambda calculus ⇒textual replacement of variable x with argument Person6.863J/9.611J Lecture 16 Sp03Another Example• S à NP VP Apply(VP, NP)• VP à V NP Apply(Apply(V, NP), NP)• Lexicon look-up Apply(Apply(lambda(o) lambda(x) (kiss past [agent x] [theme o]), NP), NP)• NP à ART N Apply(Apply(lambda(o) lambda(x) (kiss past [agent x] [theme o]),Apply(lambda (x) DEF/SING x), N)), NP)• Lexicon look-up Apply(Apply(lambda(o) lambda(x) (kiss past [agent x] [theme o]),Apply(lambda (x) (DEF/SING x), DOG)), NP)• Apply -operator Apply(Apply(lambda(o) lambda(x) (kiss past [agent x] [theme o]), (DEF/SING DOG), NP)• Apply -operator Apply(lambda(x) (kiss past [agent x] [theme DEF/SING DOG]), NP)NP à ART N Apply(lambda(x) (kiss past [agent x] [theme DEF/SING DOG]), Apply(lambda(x) (DEF/SING * x), N))• Lexicon look-up Apply(lambda(x) (kiss past [agent x] [theme DEF/SING DOG]), Apply(lambda(x) (DEF/SING * x), Person))• Apply -operator Apply(lambda(x) (kiss past [agent x] [theme DEF/SING DOG]), (DEF/SING Person))• Apply -operator (kiss past [agent DEF/SING Person] [theme DEF/SING DOG])Lexicon:V:kissed = lambda(o) lambda(x) (kiss past [agent x] [theme o])N:guy = personN:dog = DOGDet:the = DEF/SINGTop-down parse sentence:The guy kissed the dogSyntax:S à NP VPVP à V NPNP à Det NSemantics:S : Apply(VP, NP)VP : Apply(V, NP)NP : Apply(lambda (x) (DEF/SING x), N)6.863J/9.611J Lecture 16 Sp03Semantic Grammar: Definition• Syntactic and semantic processing is collapsed in a single framework• Like a regular grammar but terminal symbols are replaced by semantic categories• Example:• [VP read [NP a book]] or [write [a book]]VP à V NP⇓READ-VP à READ-VERB READ-STUFFWRITE-VP à WRITE-VERB WRITE-STUFF6.863J/9.611J Lecture 16 Sp03Example of a Grammar• RES-VP → RESERVING RES-MOD• RES-VP → RESERVING• DEP-VP → DEPARTING DEP-MODS• RESERVING → RESERVE-VERB FLIGHT• RES-MOD → for PERSON• DEPARTING → DEPART-VERB• DEPARTING → DEPART-VERB SOURCE-LOCATION• DEP-MODS→ DEP-MOD DEP-MODS• DEP-MODS→ DEP-MOD• DEP-MOD → to DEST-LOCATION• DEP-MOD → from SOURCE-LOCATION6.863J/9.611J Lecture 16 Sp03Exercise• Grammar:RES-VP à RESERVINGRES-VP à RESERVING RES-MODRESERVING à RESERVE-VERB FLIGHT-NPRES-MOD à for PERSONFLIGHT-NP à ART FLIGHT-NOUNFLIGHT-NP à ART FLIGHT-NOUN FLIGHT-MODSFLIGHT-MODS à FLIGHT-MOD FLIGHT-MODSFLIGHT-MODS à FLIGHT-MODFLIGHT-MOD à from SOURCE-LOCATIONFLIGHT-MOD à to DEST-LOCATION• Lexicon:FLIGHT-NOUN:flightART:aPERSON:meLOCATION:BostonLOCATION:ChicagoRESERVE-VERB: bookParse this sentence (bottom-up):Book a flight from Boston to Chicago for me6.863J/9.611J Lecture 16 Sp03Solution (1)• FLIGHT-MOD à from SOURCE-LOCATIONBook a flight [FLIGHT-MOD from SOURCE-LOCATION Boston] to Chicago for me• FLIGHT-MOD à to DEST-LOCATIONBook a flight [FLIGHT-MOD from SOURCE-LOCATION Boston] [FLIGHT-MOD to DEST-LOCATION Chicago] for me• FLIGHT-MODS à FLIGHT-MODBook a flight [FLIGHT-MOD from SOURCE-LOCATION Boston] [FLIGHT-MODS [FLIGHT-MOD to DEST-LOCATION Chicago]] for me• FLIGHT-MODS à FLIGHT-MOD FLIGHT-MODSBook a flight [FLIGHT-MODS [FLIGHT-MOD from SOURCE-LOCATION Boston] [FLIGHT-MODS [FLIGHT-MOD to DEST-LOCATION Chicago]]] for me6.863J/9.611J Lecture 16 Sp03Solution (2)• FLIGHT-NP à ART FLIGHT-NOUN FLIGHT-MODSBook [FLIGHT-NP ART a FLIGHT-NOUN flight [FLIGHT-MODS [FLIGHT-MOD from SOURCE-LOCATION Boston] [FLIGHT-MODS [FLIGHT-MOD to DEST-LOCATION Chicago]]]] for me• RESERVING à RESERVE-VERB FLIGHT-NP[RESERVING RESERVE-VERB Book [FLIGHT-NP ART a FLIGHT-NOUN flight [FLIGHT-MODS [FLIGHT-MOD from SOURCE-LOCATION Boston] [FLIGHT-MODS [FLIGHT-MOD to DEST-LOCATION Chicago]]]]] for me• RES-MOD à for PERSON[RESERVING RESERVE-VERB Book [FLIGHT-NP ART a FLIGHT-NOUN flight [FLIGHT-MODS [FLIGHT-MOD from SOURCE-LOCATION Boston] [FLIGHT-MODS [FLIGHT-MOD to DEST-LOCATION Chicago]]]]] [RES-MOD for PERSON me]• RES-VP à RESERVING RES-MOD[RES-VP [RESERVING RESERVE-VERB Book [FLIGHT-NP ART a FLIGHT-NOUN flight [FLIGHT-MODS [FLIGHT-MOD from SOURCE-LOCATION Boston] [FLIGHT-MODS [FLIGHT-MOD to DEST-LOCATION Chicago]]]]] [RES-MOD for PERSON me]]6.863J/9.611J Lecture 16 Sp03Useful?• Semantic grammars are useful in a limited domain• Dialogue system to book flights through the telephone• For general use à too many rules!6.863J/9.611J Lecture 16 Sp03Application Continuum• Machine Translation• Unrestricted language comprehension• Summarization• Information extraction• Find specific information: location, names of terrorists, …• Text classification• What is the text about (topic detection)?• Information retrieval6.863J/9.611J Lecture 16 Sp03Information Extraction• Analyzing unrestricted, unstructured text• Extracting specific structured information • Enabling technology• Converting text to a database (data mining)• Summarization6.863J/9.611J Lecture 16 Sp03Example from the terrorism domainInput:San Salvador, 19 Apr 89. Salvadoran President-elect Alfredo Cristiani condemned the terrorist killing of Attorney general Roberto Garcia Alvarado and accused the Farabundo Marti


View Full Document

MIT 6 863J - Lecture 16: the boundaries of syntax & semantics

Documents in this Course
N-grams

N-grams

42 pages

Semantics

Semantics

75 pages

Semantics

Semantics

82 pages

Semantics

Semantics

64 pages

Load more
Download Lecture 16: the boundaries of syntax & semantics
Our administrator received your request to download this document. We will send you the file to your email shortly.
Loading Unlocking...
Login

Join to view Lecture 16: the boundaries of syntax & semantics and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Lecture 16: the boundaries of syntax & semantics 2 2 and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?