Penn Lexical Semantics at Penn Proposition Bank and
Penn Lexical Semantics at Penn: Proposition Bank and Verb. Net Martha Palmer, Dan Gildea, Paul Kingsbury, Olga Babko-Malaya, Bert Xue, Karin Kipper, Hoa Dang, Szuting Yi, Edward Loper, Jinying Chen August 22, 2003 ARDA Visit 1
Outline Penn · Verbs and their semantic roles · The part played by word senses · Mapping Propbank sense distinctions to other sense inventories · Verb. Net entries for individual, sense tagged verbs ARDA Visit 2
Predicate-Argument Structure Penn They signed the document in spite of his objections. sign Agent: They Theme: the document NP 1[case: nom] NP 2[case: acc] Arg. M: in spite of his objections Arg 0: They REL: signed Arg 1: the document Arg. M-ADV: in spite of his objections. ARDA Visit 3
Capturing semantic roles* Penn SUBJ · Charles broke [ ARG 1 the LCD Projector. ] SUBJ · [ARG 1 The windows] were broken by the hurricane. SUBJ · [ARG 1 The vase] broke into pieces when it toppled over. *See also Framenet, http: //www. icsi. berkeley. edu/~framenet/ ARDA Visit 4
A Tree. Banked Sentence Penn (S (NP-SBJ Analysts) S (VP have (VP been VP (VP expecting (NP a GM-Jaguar pact) have VP (SBAR (WHNP-1 that) (S (NP-SBJ *T*-1) NP-SBJ been VP (VP would Analyst (VP give expecting NP s (NP the U. S. car maker) SBAR (NP an eventual (ADJP 30 %) stake) NP S (PP-LOC in (NP the British company)))))) a GM-Jaguar WHNP-1 VP pact that NP-SBJ VP *T*-1 would NP give NP Analysts have been expecting a GM-Jaguar pact that would give the U. S. car maker an the US car maker eventual 30% stake in the British company. ARDA Visit NP an eventual 30% stake PP-LOC in NP the British company 5
The same sentence, Prop. Banked Penn (S Arg 0 (NP-SBJ Analysts) (VP have been expecting (VP been (VP expecting Arg 1 (NP a GM-Jaguar pact) Arg 0 (SBAR (WHNP-1 that) (S Arg 0 (NP-SBJ *T*-1) (VP would Analyst (VP give a GM-Jaguar s Arg 2 (NP the U. S. car maker) pact Arg 1 (NP an eventual (ADJP 30 %) stake) (PP-LOC in (NP the British company)))))) Arg 0 *T*-1 that would give Arg 2 the US car maker ARDA Visit Arg 1 an eventual 30% stake in the British company expect(Analysts, GM-J pact) give(GM-J pact, US car maker, 30% stake) 6
Word Senses in Prop. Bank Penn · Orders to ignore word sense not feasible for 700+ verbs Ø Mary left the room Ø Mary left her daughter-in-law her pearls in her will Frameset leave. 01 "move away from": Arg 0: entity leaving Arg 1: place left Frameset leave. 02 "give": Arg 0: giver Arg 1: thing given Arg 2: beneficiary How do these relate to traditional word senses as in Word. Net? ARDA Visit 7
Fine-grained Word. Net Senses Penn · Senseval 2 – WSD Bakeoff, using Word. Net 1. 7 (avg polysemy: 16, ITA: 71%, best system: 59. 6%) Verb Develop WN 1: CREATE, MAKE SOMETHING NEW They developed a new technique WN 2: CREATE BY MENTAL ACT They developed a new theory of evolution develop a better way to introduce crystallography techniques? WN 1? WN 2? ARDA Visit 8
WN Senses: verb ‘develop’ WN 1 WN 2 Penn WN 3 WN 4 WN 6 WN 7 WN 8 WN 11 WN 12 WN 13 WN 19 ARDA Visit WN 5 WN 9 WN 10 WN 14 WN 20 9
Sense Groups: verb ‘develop’ Penn (Avg polysemy: 8, ITA: 82%, best system: 69%) WN 1 WN 6 WN 2 WN 3 WN 4 WN 7 WN 8 WN 11 WN 12 WN 13 WN 19 ARDA Visit WN 5 WN 9 WN 10 WN 14 WN 20 10
Prop. Bank Framesets for ‘develop’ Penn · Develop. 02 (sense: create/improve) Arg 0: agent Arg 1: thing developed Example: They developed a new technique. · Develop. 01 (sense: come about) Arg 1: non-intentional theme Example: The child developed beautifully. ARDA Visit 11
Overlap between Groups and Framesets – 95% Penn Frameset 2 Frameset 1 WN 2 WN 3 WN 4 WN 6 WN 7 WN 8 WN 11 WN 12 WN 13 WN 19 ARDA Visit WN 5 WN 9 WN 10 WN 14 WN 20 12
Sense Hierarchy Penn · Framesets – coarse grained distinctions Ø Sense Groups (Senseval-2) intermediate level (includes Levin classes) – 95% overlap § Word. Net – fine grained distinctions ARDA Visit 13
Limitations to Word. Net Penn · Poor inter-annotator agreement · Just sense tags - no representations ØVery little mapping to syntax ØNo predicate argument structure Øno selectional restrictions · No generalizations about sense distinctions ARDA Visit 14
Verb. Net Penn · Computational verb lexicon · Clear association between syntax and semantics Ø Ø Syntactic frames (LTAGs) and selectional restrictions (Word. Net) Lexical semantic information – predicate argument structure Semantic components represented as predicates Links to Word. Net senses · Entries based on refinement of Levin Classes · Inherent temporal properties represented explicitly Ø during(E), end(E), result(E) ARDA Visit 15
Verb. Net Penn Class entries: Ø Verb classes allow us to capture generalizations about verb behavior Ø Verb classes are hierarchically organized Ø Members have common semantic elements, thematic roles, syntactic frames and coherent aspect Verb entries: Ø Each verb can refer to more than one class (for different senses) Ø Each verb sense has a link to the appropriate synsets in Word. Net (but not all senses of Word. Net may be covered) Ø A verb may add more semantic information to the basic semantics of its class ARDA Visit 16
Develop. 02 “ create” – Verb. Net Penn · Levin class: grow-26. 2 , Word. Net Senses: WN 10, 12, 13, 14 · Thematic Roles: Agent[+animate] Material[+concrete] Product[+concrete] · Semantics: ¬exist(start(E), Product), exist(result(E), Product), made_of(result(E), Product, Material), cause(Agent, E) · Frames Causative/Inchoative Alternation (causative, Material Object) The gardener developed that acorn into an oak tree Causative/Inchoative Alternation (causative, Product Object) The gardener developed an oak tree from that acorn Material/Product Alternation Intransitive (Material Subject) That acorn will develop into an oak tree Material/Product Alternation Intransitive (Product Subject) An oak tree will develop from that acorn ARDA Visit 17
Develop. 01 “come about” – Verb. Net Penn · Levin Class: appear-48. 1. 1, Word. Net Senses: WN 5 · Thematic Roles : Location, Theme · Semantics: at(end(E), Theme, Location) · Frames Basic Intransitive () Intransitive (+ Location PP) Locative Inversion (Most verbs) There-insertion (Most verbs) ARDA Visit 18
Lexical Semantics at Penn · Annotation of Penn Treebank with semantic role labels (propositions) and sense tags · Links to Verb. Net and Word. Net Ø Provides additional semantic information that clearly distinguishes verb senses Ø Class based to facilitate extension to previously unseen usages ARDA Visit 19
Penn · Backup slides ARDA Visit 20
Annotation procedure Penn · Extraction of all sentences with given verb · First pass: Automatic tagging (Joseph Rosenzweig) Ø http: //www. cis. upenn. edu/~josephr/TIDES/index. html#lexicon · Second pass: Double blind hand annotation Ø ITA high 80’s to low 90’s · Third pass: adjudication Ø Tagging tool highlights inconsistencies ARDA Visit 21
Levin classes (3100 verbs) Penn • 47 top level classes, 193 second and third level • Based on pairs of syntactic frames. John broke the jar. / Jars break easily. / The jar broke. John cut the bread. / Bread cuts easily. / *The bread cut. John hit the wall. / *Walls hit easily. / *The wall hit. • Reflect underlying semantic components contact, directed motion, exertion of force, change of state • Synonyms, syntactic patterns (conative), relations ARDA Visit 22
Hit class – hit-18. 1 MEMBERS: THEMATIC ROLES: SELECT RESTRICTIONS: [bang(1, 3), bash(1), . . . hit(2, 4, 7, 10), kick (3), . . . ] Agent, Patient, Instrument Agent(int_control), Patient(concrete), Instrument(concrete) FRAMES and PREDICATES: Basic Transitive AVP cause(Agent, E) / manner (during(E), directedmotion, Agent)/ manner (end(E), forceful, Agent)/ contact(end(E), Agent, Patient) Conative AV at P manner (during (E), directedmotion, Agent) ¬contact(end(E), Agent, Patient) With/against alternation AVI against/on P cause(Agent, E) / manner(during (E), directedmotion, Instr)/ manner(end(E), forceful, Instr)/ contact (end(E), Instr, Patient)
Verb. Net/Word. Net ARDA Visit Penn 24
Action Hierarchy for Maintenance Domain ARDA Visit Penn 25
- Slides: 25