Presentation is loading. Please wait.

Presentation is loading. Please wait.

FrameNet, PropBank, VerbNet Rich Pell. FrameNet, PropBank, VerbNet  When syntactic information is not enough  Lexical databases  Annotate a natural.

Similar presentations

Presentation on theme: "FrameNet, PropBank, VerbNet Rich Pell. FrameNet, PropBank, VerbNet  When syntactic information is not enough  Lexical databases  Annotate a natural."— Presentation transcript:

1 FrameNet, PropBank, VerbNet Rich Pell

2 FrameNet, PropBank, VerbNet  When syntactic information is not enough  Lexical databases  Annotate a natural language corpus with semantic information  Largely manual classification efforts

3 Outline  FrameNet  Frame Semantics  Overview and Demo  Applications  PropBank  Overview and Demo  Applications  VerbNet  Levin Classes  Frame Demo  Conclusion

4 FrameNet

5 Frame Semantics  Sell (v) – to exchange an item for money or its equivalent  Vce (n) – estimator of reactivity delta due to voids in moderator  Definition of a word is useless without knowledge relating to that word:  Entities involved - buyer, seller, item, money  Relationships between those entities:  Buyer gives money to seller  Seller gives item to buyer  Buyer believes value of item >= monetary amount  Seller believes value of item <= monetary amount

6 Semantic Frame  Description of an event, relation, or entity and its participants  Captures the ‘essential knowledge’ of a given word sense  Developed by Charles Fillmore

7 FrameNet Overview  Attempt to represent frame semantics in a human and machine-readable database  Developed by Charles Fillmore at Berkeley’s International Computer Science Institute  Founded in 1997  Funded by National Science Foundation and DARPA  Freely available via web interface or download 

8 FrameNet Overview  Set of semantic frames  Composed of frame elements (FEs) – roles within the frame  Words that evoke this frame are called lexical units(LUs) – represent a sense of a given word  Frame: Commerce_sell  FEs: buyer, seller, item, money, place, reason…  LUs: auction.v, retail.v, vend.v…

9 Frames  Definition  Core/non-core frame elements  Definition and examples  Frame-frame relations  Lexical Units

10 Frame-Frame Relations  Inheritance – IS-A relation  Child frame is subtype of parent frame  Each frame element in parent has corresponding frame element in child  Revenge inherits from Rewards_and_punishments  Using – child frame presupposes parent frame as background  Speed presupposes Motion  No one-to-one correspondence between FEs  Subframe – child frame is subevent of complex event represented by parent  Criminal_process -> Arrest, Arraignment, Trial, Sentencing  Perspective-on – one frame provides some perspective on (perspectivizes) another frame  Commerce_goods_transfer provides perspective on Commerce_sell

11 Text Annotation  [Seller Bob] auctioned [Goods the clock] [Buyer to John]  [Item Colgate’s stock] rose [Difference $3.64][Final_value to $49.94]  reduction [Item of debt levels][Value_2 to $665 million][Value_1 from $2.6 billion]  [Sleeper They][Copula were]asleep[Duration for hours]  He took a packet of Woodbines out of the breast pocket of [Wearer his][Material cotton][Garment shirt] and lit one.

12 Development  Characterize frames  Find words that fit the frames (lexical units)  Extract sample sentences  British National Corpus (editorials, sermons, textbooks, advertisements, novels, sermons)  Linguistic Data Consortium (US newswire texts)  American National Corpus  ~200 million words  Annotate selected examples

13 Progress  1000 linked semantic frames comprising:  10,000 lexical units  170,000 manually annotated sentences  Ports to other languages  Spanish, German, Chinese, Japanese

14 Uses  Semantic role assignment  Natural language understanding  Machine translation  Part of speech tagging  Textual entailment  Information extraction  NLP applications where a syntactic parse will not suffice

15 PropBank

16  Adds a semantic layer to Penn Treebank  Attempts to capture accurate predicate-argument structure by annotating predicates and the semantic roles of their arguments  Annotates predicates (verbs) and their arguments:  John broke the window -> broke(arg0 = John, arg1 = the window)  The window broke -> broke(arg1 = the window)  Developed in 2001 at the University of Pennsylvania  Martha Palmer, Paul Kingsbury  Free, open-source, downloadable 

17 PropBank Structure  PropBank is a set of frame files  Each frame file contains one or more PropBank verb senses (aka frameset or roleset ID)  Each verb sense is annotated with:  Semantic roles for each argument of a predicate  Examples  Links to other lexical tools (FrameNet, VerbNet)

18 PropBank Arguments  Standardized as much as possible  Arg0 = agent  Arg1 = patient  Arg2 = instrument/attribute  Arg3 = starting point/attribute  Arg4 = ending point  ArgM = modifier  Obama met him privately in the White House, on Thursday.  Rel: met  Arg0: Obama  Arg1: him  ArgM-MNR: privately  ArgM-LOC: in the White House  ArgM-TMP: on Thursday

19 PropBank Example …the campaign is drawing fire from anti-smoking advocates… Arg0:the campaign Rel:drawing Arg1:fire Arg2-from:anti-smoking advocates

20 PropBank Example They-1 have *trace*-1 to sell when things look like they're falling. *trace* sell when things look like they're falling A painting by August Strindberg sold at auction in Stockholm. A painting by August Strindberg sold at auction in Stockholm

21 Differences From FrameNet  Verb-specific  Each verb is its own predicate  Closer to syntactic parse  More thorough but simpler annotation of corpus

22 PropBank Progress  3500 verbs annotated  Work on translating to Dutch, Arabic  Semantic role labeling  Knowledge discovery  Semantic parsing

23 VerbNet

24  Lexicon of English verbs  Groups verbs based upon shared syntactic behavior  5800 verbs in 270 verb classes  Based on Levin classes and their extensions  Developed by Karin Kipper-Schuler at University of Pennsylvania via NSF and DARPA grants  Free, open source, downloadable  l

25 Levin Classes  English Verb Classes and their Annotations, Beth Levin, 1993  Syntactic behavior of a verb is based upon its meaning  Possible to syntactically group verbs into classes based upon how they interact with specific objects/prepositions/subjects and expect them to have some semantic similarity  e.g. Locative alternation – involves moving something into or onto a location  Verbs of placement and covering  Scatter, pump, hang, drizzle, cram, load

26 VerbNet Roles  Groups verbs based upon Levin classes  Add semantic role labels to Levin classes, e.g.  Agent – actor in an event who carries out the event  Theme – undergoer that is central to event or state that does not have control over the way the event occurs  Destination – goal that is a concrete, physical location  …  23 total  Illustrate the “who what how when where” information contained in a sentence  Analogous to FrameNet’s frame elements or PropBank’s numbered arguments

27 VerbNet Classes  Set of member verbs  Thematic roles used in predicate-argument structure of verbs in the class  Selectional restrictions on the roles  “Sam drank a coffee.”  “Sam drank a car.”  Set of frames:  Brief description  Example  Syntactic description  Set of semantic predicates, includes temporal function indicating when a predicate is true

28 VerbNet applications  Verbs typically convey the main idea of a sentence  Maps the syntactic nature of PropBank predicate/argument parses into a richer semantic context  Machine translation  Document classification  Word sense disambiguation  Semantic role labeling  3D animation (parameterized action representations)  Planning  Automatic verb acquisition

29 Automatically Extending VerbNet  Semantic information for several verbs at a time captured in VerbNet classes  Can automatically add new candidate verbs to a class by testing against pre-defined class specifications  Removes need for exhaustive manual encodings

30 Automatically Extending VerbNet  Apply k-means clustering to some other resource:  PropBank  WordNet  FrameNet  Observe the clusters to see if they correspond to any VerbNet class  If so, do they contain any verbs not in the existing VerbNet class?  Able to add 47 verbs

31 Summary  FrameNet, PropBank, VerbNet all annotate an NL corpus with semantic information:  FrameNet – defines a set of semantic frames annotating additional semantic information needed to capture meaning of a word  PropBank – annotates propositions and their arguments in a structured fashion  VerbNet – groups verbs into syntactically and semantically similar classes  All are used when a syntactic parse is not enough  Highly linked:  Unified Verb Index -  SemLink -

32 Questions

33 References VerbNet Guidelines, Palmer, M. 2009. Semlink: Linking PropBank, VerbNet and FrameNet. Proceedings of the Generative Lexicon Conference. Sept. 2009, Pisa, Italy: GenLex-09. 2012 Data Format Specifications for English PropBank, format.txt. format.txt M. Palmer et al, “English PropBank Annotation Guidelines,” 2012, annotation-guidelines.pdf annotation-guidelines.pdf Karin Kipper, Anna Korhonen, Neville Ryant, and Martha Palmer. Extending VerbNet with Novel Verb Classes. Fifth International Conference on Language Resources and Evaluation (LREC 2006). Genoa, Italy. June, 2006. Karin Kipper, Anna Korhonen, Neville Ryant, and Martha Palmer. Extensive Classifications of English verbs. Proceedings of the 12th EURALEX International Congress. Turin, Italy. September, 2006. Paul Kingsbury and Karin Kipper. Deriving Verb-Meaning Clusters from Syntactic Structure.. Workshop on Text Meaning, held in conjunction with HLT/NAACL 2003. Edmonton, Canada, May 2003. Karin Kipper-Schuler, VerbNet: a Broad-Coverage, Comprehensive Verb Lexicon,” Dissertation, University of Pennsylvania, 2005. Michael Ellsworth et al, “FrameNet II: Extended Theory and Practice,” 2010,

Download ppt "FrameNet, PropBank, VerbNet Rich Pell. FrameNet, PropBank, VerbNet  When syntactic information is not enough  Lexical databases  Annotate a natural."

Similar presentations

Ads by Google