210 likes | 327 Vues
This announcement covers essential topics related to formal grammar and parsing techniques, focusing on top-down and bottom-up parsing methodologies within context-free grammars (CFGs). Key questions include understanding top-down parsing through examples like "Pat ate the kiwi," analyzing CFGs' effectiveness in representing English, and exploring constructions. Additionally, quizzes will assess comprehension of these concepts, including grammar rules and their applications in various parsing scenarios. Participants should prepare for a quiz due on April 22nd and review key terms related to this week's discussions.
E N D
CS 182Sections 103 - 104 Eva Mok (emok@icsi.berkeley.edu) April 21, 2004
Announcements • a9 out sometime Thursday, due Wednesday April 22nd, 11:59pm • How’s your progress on the BBS articles?
Schedule • Last Week • SHRUTI • Formal Grammar and Parsing • This Week • Construction Grammar, ECG • Psychological model of sentence processing • Next Week • Language learning • Learning ECG
Quiz • What is top-down parsing? Using a plausible CFG grammar, what is the top down parse of “Pat ate the kiwi”? • How well can CFGs represent English? What are some mechanisms for improvement? • What are constructions? • What is the difference between subcasing and evoking another schema or construction?
Quiz • What is top-down parsing? Using a plausible CFG grammar, what is the top down parse of “Harry likes the cafe”? • How well can CFGs represent English? What are some mechanisms for improvement? • What are constructions? • What is the difference between subcasing and evoking another schema or construction?
Grammar • A grammar is a set of rules defining a formal language • an example is right-regular grammar • a more common example is Context-Free Grammar β • : single non-terminal • β : any combination of terminals and non-terminals S NP VP NP Det Noun | ProperNoun VP Verb NP | Verb PP PP Preposition NP Noun kiwi | orange | store ProperNoun Pat | I Det a | an | the Verb ate | went | shop Preposition to | at
Top Down Parsing: Pat ate the kiwi • start from S and apply all applicable rules • forward search (use your favorite search algorithm…) S S NP VP NP Det Noun | ProperNoun VP Verb NP | Verb PP PP Preposition NP NP VP Det Noun VP ProperNoun VP Noun kiwi | orange | store ProperNoun Pat | I Det a | an | the Verb ate | went | shop Preposition to | at a Noun VP an Noun VP the Noun VP … a kiwi VP a orange VP a store VP succeed when you encounter Pat ate the kiwiin a state without any non-terminals (DFS)
Bottom Up Parsing: Pat ate the kiwi • start from the sentence and try to match non-teriminals to it • backward search (use your favorite search algorithm…) S S NP VP NP Det Noun | ProperNoun VP Verb NP | Verb PP PP Preposition NP NP VP NP Verb NP Noun kiwi | orange | store ProperNoun Pat | I Det a | an | the Verb ate | went | shop Preposition to | at NP Verb DetNoun NP Verb Det kiwi NP Verb the kiwi NP ate the kiwi succeed when you encounter Sin a state by itself ProperNoun ate the kiwi Pat ate the kiwi (DFS)
Quiz • What is top-down parsing? Using a plausible CFG grammar, what is the top down parse of “Harry likes the cafe”? • How well can CFGs represent English? What are some mechanisms for improvement? • What are constructions? • What is the difference between subcasing and evoking another schema or construction?
Notice the ungrammatical and/or odd sentences that we can generate? • * Pat ate a orange • *Pat shop at the store • *Pat went a store • ?Pat ate a store • ?The kiwi went to an orange S NP VP NP Det Noun | ProperNoun VP Verb NP | Verb PP PP Preposition NP Noun kiwi | orange | store ProperNoun Pat | I Det a | an | the Verb ate | went | shop Preposition to | at need to capture agreement, subcategorization, etc you could make many versions of verbs, nouns, dets cumbersome
Pat number : SG person : 3rd agreement I number : SG person : 1st agreement Went agreement Shop number : person : 1st agreement Unification Grammar • Basic idea: capture these agreement features for each non-terminal in feature structures • Enforce constraints on these features using unification rules • VP Verb NP • VP.agreement ↔ Verb.agreement • S NP VP • NP.agreement ↔ VP.agreement
Quiz • What is top-down parsing? Using a plausible CFG grammar, what is the top down parse of “Harry likes the cafe”? • How well can CFGs represent English? What are some mechanisms for improvement? • What are constructions? • What is the difference between subcasing and evoking another schema or construction?
CAFE Embodied constructions ECG Notation Form Meaning constructionHARRY form : /hEriy/ meaning : Harry Harry constructionCAFE form : /khaefej/ meaning : Cafe cafe Constructions have form and meaning poles that are subject to type constraints.
Quiz • What is top-down parsing? Using a plausible CFG grammar, what is the top down parse of “Harry likes the cafe”? • How well can CFGs represent English? What are some mechanisms for improvement? • What are constructions? • What is the difference between subcasing and evoking another schema or construction?
A schema hierarchy of objects (Nomi) schema Entity schema Physical-Object subcase of Object, Place schema Human subcase of Animate roles sex schema Place schema Animate subcase of Physical-Object roles animacy constraints animacy ← true schema Object subcase of Entity schema Nomi subcase of Human sex ← female slot filler schema Referent subcase of Entity roles category distribution boundedness number gender accessibility resolved-ref schema Toy subcase of Manipulable-Object schema Manipulable-Object subcase of Physical-Object schema Cup subcase of Manipulable- Object schema Ball subcase of Toy
The schemas we just defined Entity Place Referent Object Physical-Object Food Manipulable-Object Animate Cup Toy Human Cake Juice Ball Book Block Doll Bed Nomi Mother Father
A schema hierarchy of actions (Nomi) schema Action roles agent : Entity schema CauseMove subcase of DirectedAction, Move roles causer : Human mover : Physical-Object motion : Move constraints motion.mover ↔ mover motion.agent ↔ causer motion.direction ↔ direction agent ↔ causer patient ↔ mover type constraint schema DirectedAction subcase of Action roles patient : Entity identification constraint schema Move subcase of Action roles mover : Entity direction : Place
The schemas we just defined Action Directed Action Move Cause-Move Transfer JointMove
Constructions, finally (Nomi) construction Ref-Expr form : Schematic-Form meaning : Referent construction Cup-Cn level 0 subcase of Ref-Expr form : Word self.f.orth ← “cup" meaning evokes Cup as n self.m.category ↔ n self.m.resolved-ref ↔ n construction Nomi-Cn level 0 subcase of Ref-Expr form : Word self.f.orth ← "Nomi" meaning evokes Nomi as n self.m.category ↔ n self.m.resolved-ref ↔ n local name fancy way of saying that the category of the referent is Nomi
Constructions, finally (Nomi) construction Motion-Verb meaning : Move construction Cause-Motion-Verb subcase of Motion-Verb meaning : CauseMove construction Get-Cn level 0 subcase of Cause-Motion-Verb form : Word self.f.orth ← "get" lexical construction
Constructions, finally (Nomi) construction Transitive-Cn level 2 constructional constituents agt : Ref-Expr v : Cause-Motion-Verb obj : Ref-Expr form agt.f before v.f v.f before obj.f meaning v.m.agent ↔ agt.m.resolved-ref v.m.patient ↔ obj.m.resolved-ref smaller constructions that it takes ordering constraints on the constituents