Out line Communicat ion Symbolic Nat ur al Language Processing - PDF document

Out line • Communicat ion • Symbolic Nat ur al Language Processing Communicat ion • Reading: R&N Sect . 22.1-22.6 J uly 14, 2005 CS 486/ 686 Univer sit y of Wat erloo 2 CS486/686 Lecture Slides (c) 2005 P. Poupart Communicat ion Turing Test • Communicat ion: int ent ional exchange of • Can a comput er f ool a human t o t hink inf ormat ion brought about by t he t hat it is communicat ing wit h anot her product ion and percept ion of signs human? drawn f rom shared syst em of convent ion. • Language: – Enables us t o communicat e – I nt imat ely t ied t o t hinking 3 4 CS486/686 Lecture Slides (c) 2005 P. Poupart CS486/686 Lecture Slides (c) 2005 P. Poupart Speech Component s of Communicat ion • Speech: communicat ion act • I nt ent ion – Speaker S decides t hat t here is some proposit ion P – Talking wort h saying t o hearer H. – Writ ing ut t erances • Generat ion – Facial expression – Speaker plans how t o t urn proposit ion P int o an – Gest ure ut t erance (i.e. a sequence of words W) • Synt hesis situation – Speaker produces t he physical realizat ion W’ of t he words W (i.e., vibrat ion in air, ink on paper) ut t erances Speaker Hearer 5 6 CS486/686 Lecture Slides (c) 2005 P. Poupart CS486/686 Lecture Slides (c) 2005 P. Poupart 1

Component s of Communicat ion Component s of Communicat ion • Percept ion • Disambiguat ion – Hearer perceives physical realizat ion W’ as W 2 and – Hearer inf ers t hat speaker int ended t o decodes it as t he words W 2 (i.e., speech convey P i (where ideally P i = P). recognit ion, opt ical charact er recognit ion) • I ncorporat ion • Analysis – Hearer decides t o believe P i (or not ). – Hearer inf ers W 2 has possible meanings P 1 , P 2 , … , P n – Three part s: • Synt act ic int erpret at ion • Semant ic int erpret at ion • Pragmat ic int erpret at ion 7 8 CS486/686 Lecture Slides (c) 2005 P. Poupart CS486/686 Lecture Slides (c) 2005 P. Poupart Component s of Communicat ion Dif f icult ies SPEAKER Synthesis: • How could communicat ion go wrong? Intention: Generation: [thaxwahmpaxsihzdehd] Know(H, Alive(Wumpus,S )) "The wumpus is dead" L 3 – I nsincer it y – Speech recognit ion err ors HEARER – Ambiguous ut t erance Perception: Analysis: Disambiguation: S (Parsing): – Dif f erent cont ext s "The wumpus is dead" L Alive(Wumpus,S ) NP VP 3 Article Noun Verb Adjective The wumpus is dead Incorporation: (Semantic Interpretation): L Alive(Wumpus,Now) Tired(Wumpus,Now) TELL( KB, L Alive(Wumpus,S ) (Pragmatic Interpretation): Alive(Wumpus,S ) 3 L 3 Tired(Wumpus,S ) 3 9 10 CS486/686 Lecture Slides (c) 2005 P. Poupart CS486/686 Lecture Slides (c) 2005 P. Poupart Language Grammar • For mal language • Grammar specif ies t he composit ional st r uct ur e of complex messages – Set of st rings of t erminal symbols (words) – St rict rules • Each st r ing in a language can be analyzed/ generat ed by t he grammar – E.g., f irst order logic, J ava • A grammar is a set of rewrit e rules • Nat ur al language – S � NP VP – No st rict def init ion – Art icle � t he | a | an | … – Chinese, Danish, English, et c. 11 12 CS486/686 Lecture Slides (c) 2005 P. Poupart CS486/686 Lecture Slides (c) 2005 P. Poupart 2

Grammar Types Grammar Types • Regular grammar: • Cont ext sensit ive grammar: – nont erminal � t erminal [nont er minal] – More t erminals on right -hand side – S � a S – ASB � AAaBB – S � b • Recursively enumerable grammar: • Cont ext f ree grammar (CFG): – No const raint s – nont erminal � anyt hing – S � aSb 13 14 CS486/686 Lecture Slides (c) 2005 P. Poupart CS486/686 Lecture Slides (c) 2005 P. Poupart Lexicon example Grammar example • Noun � breeze | glitter | agent • S � NP VP | S Conjunction S • Verb � is | see | smell | shoot • NP � Pronoun | Name | Noun | Article Noun | NP PP | NP RelClause • Adj ect ive � right | lef t | east | dead • VP � Verb | VP NP | VP Adjective | VP • Adverb � there | nearby | ahead PP | VP Adverb • Pronoun � me | you | I | it • PP � � Preposition PP � � • Name � John | Mary | Boston • RelClause � � that VP • Art icle � the | a | an � � 15 16 CS486/686 Lecture Slides (c) 2005 P. Poupart CS486/686 Lecture Slides (c) 2005 P. Poupart Grammat icalit y J udgement s Grammat icalit y J udgement s Set of strings • Overgenerat ion examples: – Me go Bost on. Natural Grammar – I smell pit gold wumpus not hing east . language agreement • Undergener at ion example: – I t hink t he wumpus is smelly Goal: design grammar to match natural language 17 18 CS486/686 Lecture Slides (c) 2005 P. Poupart CS486/686 Lecture Slides (c) 2005 P. Poupart 3

Synt act ic Analysis Top-down parsing • Parsing: process of f inding a par se t ree • St art wit h S and search f or a t ree t hat f or a given input st ring has st r ing at leaves S S NP NP VP VP NP NP proposition proposition noun noun pronoun verb pronoun verb I shoot the wumpus I shoot the wumpus 19 20 CS486/686 Lecture Slides (c) 2005 P. Poupart CS486/686 Lecture Slides (c) 2005 P. Poupart Bot t om up parsing Parsing ef f iciency • St art wit h st r ing and search f or a t ree • Top-down and bot t om up parsing t hat has S as root inef f icient … – Exponent ial running t ime S • Alt ernat ive: chart par sing NP VP NP – Dynamic progr amming proposition noun pronoun verb – Cubic running t ime I shoot the wumpus 21 22 CS486/686 Lecture Slides (c) 2005 P. Poupart CS486/686 Lecture Slides (c) 2005 P. Poupart Augment ed Grammars Parse ambiguit y • Grammars t end t o overgenerat e • Some sent ences have many grammat ical – Ex: “me eat apple” parses • Augment gr ammar t o requir e • Example: – Agreement bet ween subj ect and verb – “Fall leaves f all and spring leaves spring” • Ex: “I smells” vs “I smell” – Agreement bet ween verb subcat egory and complement • Ex: “give t he gold t o me” • Ex: “give me t he gold” 23 24 CS486/686 Lecture Slides (c) 2005 P. Poupart CS486/686 Lecture Slides (c) 2005 P. Poupart 4

Semant ic I nt erpret at ion Ambiguit y • Ext ract meaning f rom ut t erances • Possible causes: • Tradit ional approach – Met onymy: f igure of speech in which one obj ect is used t o st and f or anot her – Express meaning wit h logic – Met aphor: f igure of speech in which a phrase wit h one lit eral meaning is used t o • Problem suggest a dif f erent meaning by analogy – Ambiguous semant ics – Vagueness – Ex: “Helicopt er powered by human f lies” – Unknown cont ext 25 26 CS486/686 Lecture Slides (c) 2005 P. Poupart CS486/686 Lecture Slides (c) 2005 P. Poupart Next Class Cont ext / Experience • Meaning of t en grounded in experience • Next Class: •Probabilist ic Language Processing •Russell and Norvig Ch. 23 • But humans and machines have dif f er ent experiences because of dif f erent sensors… • I s t hat a problem f or nat ur al language under st anding? 27 28 CS486/686 Lecture Slides (c) 2005 P. Poupart CS486/686 Lecture Slides (c) 2005 P. Poupart 5