
Connecting WordNet, FrameNet and VerbNet together Nadiya Yampolksa SS2007 University of Saarland Seminar: Resources for Computational Linguists Magdalena Wolska and Michaela Regneri Outline WordNet VerbNet FrameNet (already covered) Combining Resources: “Putting Pieces Together: Combining FrameNet, VerbNet and WordNet for Robust Semantic Parsing”. Lei Shi and Rada Mihalcea. Combining Resources: why and how? Combining Resources: Algorithms Connecting VerbNet to WordNet Connecting FrameNet to VerbNet Labeling VerbNet Syntactic Frame Arguments with Semantic Roles Applications which use rich lexical resources Outline WordNet. Princeton University. • WordNet • VerbNet http://wordnet.princeton.edu/ • Combining Resources: Synsets = synonym sets Algorithms WordNet 2.0 has a network of 152,046 • Applications words WordNet includes semantic relations across concepts (more than 250.000 relations in WN 2.0): . hypernymy/hyponymy, meronymy/holonymy, antonymy, entailment, etc. Download: http://wordnet.princeton.edu/obtain Take a glance at standalone application Outline • WordNet How to use in a different program? • VerbNet http://wordnet.princeton.edu/links : scripts • Combining Resources: in different programming languages for Algorithms playing around with WN. • Applications Ex.: A Perl extension module for accessing and manipulating WordNet (by Dan Brian). This module allows access to the Wordnet lexicon from Perl applications, as well as manipulation and extension of the lexicon. Texts semantically annotated with WordNet 1.6 senses (created at Princeton University), and automatically mapped to WordNet 1.7 and WordNet 1.7.1: http://lit.csci.unt.edu/index.php/Downloads#W ordNet_mappings Tagged fragment of Brown corpus Brown Corpus semantically tagged fragment: • WordNet <wf cmd=done pos=NN lemma=september wnsn=1 • VerbNet lexsn=1:28:00::>September</wf> • Combination <wf cmd=done pos=NN lemma=october wnsn=1 • Applications lexsn=1:28:00::>October</wf> <wf cmd=done pos=NN lemma=term wnsn=2 lexsn=1:28:00::>term</wf> <wf cmd=done pos=NN lemma=jury wnsn=1 lexsn=1:14:00::>jury</wf> WordNet entries for term: The noun term has 7 senses (first 5 from tagged texts) 1. (307) term -- (a word or expression used for some particular thing; "he learned many medical terms") 2. (216) term -- (a limited period of time; "a prison term"; "he left school before the end of term") 3. (113) condition, term -- ((usually plural) a statement of what is required as part of an agreement; "the contract set out the conditions of the lease"; "the terms of the treaty were generous") … …augmented with lexical file information Brown Corpus semantically tagged fragment: • WordNet <wf cmd=done pos=NN lemma=september wnsn=1 • VerbNet lexsn=1:28:00::>September</wf> • Combination <wf cmd=done pos=NN lemma=october wnsn=1 • Applications lexsn=1:28:00::>October</wf> <wf cmd=done pos=NN lemma=term wnsn=2 lexsn=1:28:00::>term</wf> <wf cmd=done pos=NN lemma=jury wnsn=1 lexsn=1:14:00::>jury</wf> WordNet entries for term: The noun term has 7 senses (first 5 from tagged texts) 1. (307) <noun.communication> term -- (a word or expression used for some particular thing; "he learned many medical terms") 2. (216) <noun.time> term -- (a limited period of time; "a prison term"; "he left school before the end of term") 3. (113) <noun.communication> condition2, term2 -- ((usually plural) a statement of what is required as part of an agreement; "the contract set out the conditions of the lease"; "the terms of the treaty were generous") … …augmented with location in the database Brown Corpus semantically tagged fragment: • WordNet <wf cmd=done pos=NN lemma=september wnsn=1 • VerbNet lexsn=1:28:00::>September</wf> • Combination <wf cmd=done pos=NN lemma=october wnsn=1 • Applications lexsn=1:28:00::>October</wf> <wf cmd=done pos=NN lemma=term wnsn=2 lexsn=1:28:00::>term</wf> <wf cmd=done pos=NN lemma=jury wnsn=1 lexsn=1:14:00::>jury</wf> WordNet entries for term: The noun term has 7 senses (first 5 from tagged texts) 1. (307) {06220694} <noun.communication> term -- (a word or expression used for some particular thing; "he learned many medical terms") 2. (216) {15025298} <noun.time> term -- (a limited period of time; "a prison term"; "he left school before the end of term") 3. (113) {06680100} <noun.communication> condition2, term2 -- ((usually plural) a statement of what is required as part of an agreement; "the contract set out the conditions of the lease"; "the terms of the treaty were generous") … … and finally augmented with sense number Brown Corpus semantically tagged fragment: • WordNet <wf cmd=done pos=NN lemma=september wnsn=1 • VerbNet lexsn=1:28:00::>September</wf> • Combination <wf cmd=done pos=NN lemma=october wnsn=1 • Applications lexsn=1:28:00::>October</wf> <wf cmd=done pos=NN lemma=term wnsn=2 lexsn=1:28:00::>term</wf> <wf cmd=done pos=NN lemma=jury wnsn=1 lexsn=1:14:00::>jury</wf> WordNet entries for term: The noun term has 7 senses (first 5 from tagged texts) 1. (307) {06220694} <noun.communication> term#1 -- (a word or expression used for some particular thing; "he learned many medical terms") 2. (216) {15025298} <noun.time> term#2 -- (a limited period of time; "a prison term"; "he left school before the end of term") 3. (113) {06680100} <noun.communication> condition2#4, term2#3 -- ((usually plural) a statement of what is required as part of an agreement; "the contract set out the conditions of the lease"; "the terms of the treaty were generous") … Outline • WordNet VerbNet. University of Colorado at Boulder – • VerbNet verb lexicon with explicitly stated syntactic and • Combining Resources: semantic info based on Levin’s verb classification Algorithms http://verbs.colorado.edu/~mpalmer/projects/v • Applications erbnet.html Download: http://verbs.colorado.edu/~mpalmer/projects/v erbnet/downloads.html 239 verb structures in VerbNet 2.1 in XML format Glance at sample entry VerbNet 2.1 sample accompany-51.7.xml • WordNet • VerbNet dedicate-79.xml • Combining Resources: Algorithms Inspector (Java app): easy retrieval of only • Applications necessary information about the verb. http://verbs.colorado.edu/verb-index/inspector/ w - WordNet sense tags (-Vm required) t - thematic roles u - selectional restrictions for thematic roles (-Vt required) r - frames e - examples x - syntax z - selectional restrictions for syntax (-Vx required) s - semantics Outline • WordNet Combining resources: • VerbNet What features of component resources makes it • Combining so useful and doable Resources: Algorithms Algorithms of combining resources (based on Lei • Applications Shi and Rada Mihalcea) Underlying grounds • WordNet FrameNet: each annotated sentences exemplifies • VerbNet possible syntactic realization for the semantic roles • Combining associated with a frame for a given target word Resources: Algorithms (here: verbs only) • Applications VerbNet: same VerbNet class (Levin) share common syntactic frame, therefore, have the same behavior WordNet: related meanings in the WN hierarchy Pluses and Minuses • WordNet FrameNet • VerbNet + good generalization across predicates using • Combining frames and semantic roles Resources: Algorithms - does not define selectional restrictions for • Applications semantic roles; limited coverage VerbNet + better coverage; defines syntactic-semantic relations - thematic roles are too generic WordNet + almost complete coverage of English verbs augmented with relational information b/w verb senses - does not encode synt. or sem. behaviour (pred- argument realization) How to profit from all • WordNet Goal 1. augment the frame semantics with VerbNet • VerbNet verb classes by labeling FrameNet frames and • Combining semantic roles with VerbNet verb entries and Resources: Algorithms corresponding arguments; • Applications Goal 2. extend the coverage of FrameNet verbs by exploiting both VerbNet verb classes and WordNet verb synonym and hyponym relations; Goal 3. identify explicit connections between semantic roles and semantic classes, by encoding selectional restrictions for semantic roles using WordNet noun hierarchies. VN/FN to WN: defining selectional restrictions • WordNet VN lexical entries are already linked to WN. Now we • VerbNet want to also link selectional restrictions to noun • Combining classes in WN Resources: Algorithms VerbNet selectional restrictions are specified as • Applications generic terms (person, concrete, instrument, etc.) Map the restrictions to ontological classes, as indicated in WN (semantic hierarchy of nouns) Example: instrument (VN) instrumentality (WN) semi-automatic ___________________________________________ FN lexical units get attached with a list of sense IDs in WN similar to annotation of VN in WN allows to see the direct mapping of VN and FM manual annotation Mapping: FN verb senses with WN 2.0 3.094 entries (http://mira.csci.unt.edu/~stone/home.html) … closure open open%2:35:00:: closure close close%2:35:00:: closure lace lace%2:35:01:: judgment boo boo%2:32:00:: judgment fault fault%2:32:00:: judgment disapprove disapprove%2:31:00:: disapprove%2:32:00:: judgment scorn scorn%2:37:00:: scorn%2:32:00:: judgment mock mock%2:32:00:: mock%2:32:01:: … Connecting FN to VN • WordNet STEP 1. Mapping VN verb entries to appropriate • VerbNet semantic frames in FN; • Combining STEP 2. Linking arguments of VN syntactic frames with Resources: Algorithms corresponding FN semantic roles. • Applications Pre-process: Divide all VerbNet verb entries
Details
-
File Typepdf
-
Upload Time-
-
Content LanguagesEnglish
-
Upload UserAnonymous/Not logged-in
-
File Pages27 Page
-
File Size-